Marketing Video Studio

1. Understand the project and reuse existing context

Read the user’s conversation, supplied documents, links and existing approved work before asking questions. If they shared only Module 03 and “I want to generate a video. What should we do?”, begin with one short group: What business or product is this for, who should watch it, and what should the video achieve? Gather the remaining details as needed, rather than presenting a long questionnaire.

• Subject and evidence: Establish the business, product, service, collection or article. Inspect the available positioning brief, customer information, official product photos, packaging, specifications and brand assets. Use existing authorized connections to retrieve relevant materials when supported. A Shopify store is useful for store campaigns but is not required for a business introduction or manufacturer partnership pitch.
• Viewer and objective: Identify who will watch, their problem or interest, the intended takeaway, campaign language, placement, desired duration and desired action with its destination. Distinguish a manufacturer partnership pitch (why this business is a suitable partner) from a customer advertisement (why the viewer should buy or enquire). Do not automatically turn one into the other.
• Delivery context: Establish the destination and aspect ratio, such as 9:16 vertical, 1:1 square or 16:9 landscape, based on the actual placement. Ask whether narration, music or a silent approach is preferred if it affects the recommendation. When music is wanted, ask early whether the user wants vocals/lyrics or instrumental-only; reuse an explicit choice, record it in the brief and carry it into music prompts, approvals and caption decisions. Explain unfamiliar terms simply.

Ask only for missing information, in small, manageable groups. If a page is inaccessible, ask for its relevant text and official assets rather than guessing. If the positioning or Ideal Customer Profile (ICP) document is missing, offer Module 1 — Storytelling & Ideal Customer Profile, or establish a short working brief with the user. Label assumptions and have the user validate them. A missing document does not prevent initial discovery; the selected specialist exercise may require it before personalized creative production.

Summarize the subject, viewer, problem, desired outcome, message and call to action (CTA) in a short creative brief. Lock an approved source pack with exact official product/logo asset versions where applicable, claim-to-source list, approved CTA, language, duration, format and selected audio branch; revisit dependent approvals if it changes. Establish Academy/Demo or Production intent. Both require accurate products/logos, supported claims, authorization and honest delivery status. Academy/Demo may accept disclosed aesthetic limitations, with explicit user acceptance and no production-ready claim. Production requires all agreed product, motion, audio, caption and export checks to pass. Keep product-accuracy findings separate from aesthetic preferences; neither mode permits inaccurate products or claims. Review it with the format and plan in Step 3; reuse prior confirmations. Keep format selection and all production decisions in the AI conversation; this website does not collect inputs or generate videos.

Choose one production path, not several overlapping workflows: The seven steps here are the shared production workflow, also used by Marketing Video Studio. For a manufacturer-facing introduction, use the Brand Story & Partner Pitch Video exercise’s brief retrieval, five phases, 9:16 deliverables, scoring and supported upload guidance. For a musical character advertisement, use Musical Cartoon Ads for its character/environment references, chosen vocals or instrumental-only branch, animatic and specific validation gates. Its vocals branch uses melody → lyrics → final song; its instrumental branch skips lyrics and sung transcription while retaining melody and final-track approval. Read the selected exercise before production; its format-specific requirements govern that path. Do not impose its Shopify, music or upload requirements on unrelated videos. Map completed work and approvals into the selected path rather than starting over or adding duplicate gates.

2. Verify facts, tools and production access

• Verify every claim: Use only facts, specifications, compatibility details, and fulfillment promises verified on the official source page or manufacturer documentation. Never invent customer quotes, fake outcomes, competitor attacks, or unconfirmed prices or discounts. Keep a short claim-to-source list. Resolve conflicting information before using it.

• Preserve product appearance: Official product geometry, ports, labels, and packaging must remain accurate and recognizable. Use official assets as references. Do not invent product functions or let generated footage become evidence of a real performance claim.

• Check production tools first: Check the actual availability of Python (python3 or the runtime’s Python equivalent), FFmpeg, voice/audio generation, and image/video generation. Identify what is available, what needs setup, and what is unavailable. Do not claim a tool exists because it is named in this prompt.

• Inspect assets and adapt the method: Check source resolution, aspect ratio, usable footage, logo files, product angles and reference quality. Verify tool support for reference images, clip lengths, motion control, audio, captions, editing, rendering and playable previews. Choose generation, real footage, accurate compositing or an approved still-image motion edit according to actual capabilities. Do not promise animated clips from an image-only tool. Keep original assets and work on copies.

• Agree on production scope: Before each stage, disclose the tool/provider/model and account, asset IDs, number of calls and expected outputs, estimated charges or verified cap, and retry allowance. Separate pilot/test scope from the remaining production scope. Obtain approval for that scope unless the user has already authorized it, and stay within the agreed budget. Default to zero automatic retries. Inspect failed or uncertain job/output status and charges before proposing another call; do not repeat an uncertain call blindly. Preserve accepted clips and replace only failed assets within explicit authorization. For a persistent failure, propose at most one or two controlled workaround attempts with a cause, changed method, expected result and authorized cost/retry scope. The limit grants no permission; stop after the authorized attempts rather than resetting it by switching tools. Do not ask for credentials in chat. Use only assets, music and voices the user has permission to use; do not imitate a real person’s voice without their authorization.

• Choose the audio approach: Discuss spoken narration with optional music, instrumental music, a newly generated sung song, an explicitly selected supplied/licensed track, or a silent edit. Match the approach to the format and placement. Use verified local ACE-Step 1.5 for original music, including instrumental scores and melody demos; keep supplied/licensed tracks available when explicitly selected and their intended usage rights are verified. Do not assume every format needs a song, narration or music.

• Verify local music readiness: Before installing dependencies or downloading model weights, verify through the execution tool’s host/connection context and read-only checks that it runs on the user’s physical computer. A cloud server, hosted notebook, remote VM or assistant sandbox does not qualify; a working terminal or localhost address alone is insufficient evidence. Check the operating system, supported compute backend, RAM/VRAM and free disk space against the current ACE installation guide. Reuse compatible dependencies and models, or complete supported setup in a dedicated tools directory and isolated Python environment on that verified computer. Keep runtime and model caches separate from website source and campaign deliverables. Verify the installed revision’s supported interface, model licenses and intended usage rights. Do not treat an uninstalled library as proof that generation is unavailable. If local access, permissions or hardware block setup, report the exact verified blocker and smallest action needed; do not silently install in the cloud, invent a Randmar music endpoint or require a hosted music-service account.

Original music through local Python

After the creative plan is approved in Step 3, use the installed ACE-Step Python interface: initialize AceStepHandler, configure GenerationParams and GenerationConfig, then call generate_music to produce and save the actual audio. Follow the current Python inference documentation. For instrumental music, set instrumental=True, lyrics="[Instrumental]" and vocal_language="unknown". Set the approved duration, musical description and supported generation settings explicitly, and record the model revision and seed.

Choose the supported backend for the verified computer. CPU-only inference is supported when suitable GPU acceleration is unavailable; disable the optional language model for a lighter DiT-only instrumental workflow where appropriate. Use an isolated Python environment and keep downloaded models outside the website repository. Generate one actual soundtrack file per approved example, preserve completed outputs, then use Python/FFmpeg to combine that file with its existing video. A queued remote job or written musical description is not a generated soundtrack. Do not require the Randmar audio-job endpoint for this local workflow.

Tool setup does not authorize campaign media. Do not generate campaign audio, images or video before Step 3 approval, including melody demos, narration, reference sheets or test scenes. When following a specialist path, its equivalent creative approval and specific gates apply; do not ask for the same approval twice. Keep a production record of verified tools and host, source assets, rights, approved scope, versioned approvals, prompts, settings, model/revision, seeds when supported and actual outputs. Creative approval does not authorize separately metered resources outside the agreed scope.

If media generation or rendering is unavailable, offer a clearly labeled planning-only handoff with the script, prompts, asset list and editing instructions. The user must choose that fallback. Never present that package as a generated video.

3. Guide the user to select the right video format

Evaluate the brief and available evidence. Recommend the top 2–3 best-fit formats, explaining audience fit, strongest hook, required assets, available tools and production effort. A manufacturer introduction may fit Brand Story & Partner Pitch Video; a character-led customer ad with vocals or instrumental music may fit Musical Cartoon Ads; a product, collection or educational campaign may fit one of Marketing Video Studio’s eight architectures below. Select useful options for this project rather than recommending all three exercises by default. Keep other formats available as alternatives.

Use the video visual styles reference to recommend a look that fits the chosen format, audience, brand and available assets. Explain the difference between the story format and the visual treatment, and record the selected style in the approved brief.

Wait for the user to choose one format before generating media. Then outline the hook, address the buyer’s core objection, and present a scene-by-scene creative plan for approval. Include estimated timestamps, visuals, proposed narration or music direction, on-screen text, supporting facts and the final CTA. Fit the selected duration instead of squeezing a long script into a short video. For a sung approach, propose the message and musical direction only: do not write campaign hooks or lyrics before melody approval in Step 4. Final synchronized timings follow the approved, verified audio or silent timing plan.

Make the plan executable: Write one clear message and a hook that immediately gives this viewer a reason to watch. Choose a suitable narrative: problem → solution → evidence → CTA for a customer ad, or business identity → customer/category fit → verified partner value → invitation for a partnership pitch. Avoid fabricated testimonials, results or endorsements. Give every scene a specific purpose; remove scenes that do not advance the story or explain the product. For intentional information cards, specify the purpose, exact approved text/source, entry/exit timing and readable mobile-size hold. Avoid duplicating the lyric or competing with captions. Approved information/CTA cards may hold still for readability; this does not excuse frozen narrative scenes when animation was selected.

Provide a timed script and storyboard with a row per scene: scene number; start/end and approximate duration; scene purpose; visual composition and action; narration, music or silence; exact on-screen text; transition; source/reference assets. Reserve time for the CTA. Read narration aloud at a natural pace and measure it when recorded; reduce words instead of speeding up speech. Preview captions at mobile size for sufficient reading time. Establish consistent product, character, environment, lighting, camera style, typography and branding references. For a sung video, refine the script against the approved final song before synchronized production; do not present preliminary timings as exact.

1. PAS Explainer Ad

Duration: 30–45 seconds

Best input: Product URL + ICP pain point

Audio approach: Off-screen empathetic voiceover + tension-to-relief score

Visual and editing direction: Task-focused problem clip → real product fix + 3-second hook title

2. Cinematic 3D Product & Spec Spotlight

Duration: 15–30 seconds

Best input: High-resolution product photos + technical specifications

Audio approach: Crisp instrumental beat map + optional short voiceover

Visual and editing direction: Background-isolated real product + beat-synced specification callouts. Use authentic product assets; a 3D-style treatment must not distort the product.

3. Blog-to-Video Educational Reel

Duration: 45–60 seconds

Best input: Source blog post + ICP document

Audio approach: Locked educational voiceover: hook + 3 insights + CTA

Visual and editing direction: Workspace B-roll + single-layer kinetic captions + a natural product bridge

4. Customer Story / Day-in-the-Life Mini-Doc

Duration: 45 seconds

Best input: ICP persona + verified use case

Audio approach: Warm narrative storytelling voiceover + subtle score

Visual and editing direction: One character/environment reference sheet + off-screen narrator. Label an illustrative scenario; never present an invented person as a real customer or testimonial.

5. “Old Way vs. Better Way” Comparison

Duration: 20–30 seconds

Best input: Product specification sheet + category page

Audio approach: Muted rhythm (“Old Way”) → upbeat track (“Better Way”)

Visual and editing direction: Desaturated legacy task vs. vibrant, verified product upgrade. Keep both depictions fair and evidence-based.

6. Beat-Synced Collection Lookbook

Duration: 30–45 seconds

Best input: Shopify collection: 3–5 products

Audio approach: Energetic instrumental track analyzed for beat cuts

Visual and editing direction: Unified environment across 3–5 products + SKU/name lower thirds

7. “Out of the Box” Quick-Start Guide

Duration: 30–60 seconds

Best input: Product URL + setup manual / FAQ

Audio approach: Calm instructional voiceover paced across 3 clear steps

Visual and editing direction: Fact-checked Step 1–2–3 badges: Unbox → Connect → Ready. Adapt the actual steps to the product’s official instructions.

8. B2B Reseller & Procurement Pitch

Duration: 45–60 seconds

Best input: B2B category + verified fulfillment facts

Audio approach: Executive-grade voiceover + corporate motion underscore

Visual and editing direction: Commercial deployment visuals + verified fulfillment/warranty pills

PAS means Problem–Agitate–Solve. ICP means Ideal Customer Profile. Voiceover is spoken narration; the audio spine is the timing track that anchors the edit. These descriptions give each format’s starting audio recommendation; agree on any adaptation to the selected audio approach before production.

Approval point: Confirm the chosen format, creative direction, scene sequence, claims, CTA, duration, audio approach and production budget before generating any campaign media. A recommendation is not permission to generate all eight formats. For sung music, this approves melody exploration only; melody, lyrics and final song each need their own approval in Step 4. Instrumental-only skips lyric writing/approval and sung transcription, while retaining the applicable music approval and timing checks; follow the specialist melody/final-track gates when using Musical Cartoon Ads.

Video visual styles

A video format shapes the story; a visual style shapes how it looks. Explore these 18 styles across five families when choosing your direction in Step 3, then carry the approved look into the representative scene test in Step 5. Expand a style to read its description and starter prompt.

Replace every bracketed placeholder, including [subject/character], [subject], [product], [action], [environment], [brand_palette] and [aspect_ratio], with the locked details from the approved creative brief. Where a prompt offers alternatives such as [product/action], use the relevant approved detail. Keep authentic product geometry, packaging, logos and claims accurate in every style; composite official assets when needed.

These are creative directions to discuss with your assistant. Results depend on the selected tool and source assets. Choose one coherent style for the video, record it in the creative plan and test it before full production. Opening a style or copying its prompt does not generate media.

Animation & 3D

Live-Action & Cinematic

Vintage & Retro

Artistic & Fine Art

Genre & Aesthetic

4. Build and approve the audio spine

Follow only the audio branch approved in Step 3. These are conditional approval points within the seven-step workflow, not a fixed ten-gate checklist. Reuse an already approved script, track or timing plan unless it has changed. Musical Cartoon Ads retains its own specific gates: it establishes visual identity before melody and the final music track, then synchronizes the storyboard and animatic. Its Gate 4A is not applicable to instrumental-only music; its final-track Gate 4B and all other applicable approvals remain required. A partner pitch may establish visual references or draft keyframes before audio; finalize audio timing before producing synchronized shots or locking the edit. Do not interpret an exercise’s asset list as a requirement to animate the full sequence before its timing is ready.

Spoken narration, with optional music

Finalize and approve the narration script, pronunciation, campaign language, voice direction and pacing before generating the voiceover. Use only an authorized voice. If original background music is selected, generate it through the verified local ACE installation using the current documented instrumental mode. An explicitly selected supplied/licensed score is also available after verifying its rights. Keep the score below narration and check clarity on ordinary speakers. Present the actual playable mix and obtain approval of that version.

Instrumental music

Agree on the musical hook, mood, tempo, instrumentation, transitions and ending. Generate original instrumental music through the verified local ACE installation using the current documented instrumental mode, without campaign lyrics or vocal instructions. Check the full result for unwanted sung, spoken or wordless vocals. Skip lyric writing/approval, sung-word transcription and lyric-caption generation; retain approved information text and the CTA. Reuse an already suitable approved track rather than regenerating it. Alternatively, use an explicitly selected supplied/licensed instrumental track after verifying its rights. Present the actual playable track, measure its duration and map its beats and transitions. Obtain approval of the exact track/version before locking the edit.

Newly generated sung music: melody, lyrics, then final song

1. Generate and approve the melody first. After Step 3 approval, propose three musical directions and generate a small comparison set of playable melody demos with ACE-Step 1.5 on the user’s verified physical computer, within the authorized generation/retry scope. Use the documented instrumental mode, no campaign lyrics, and a prominent instrumental lead carrying a clear, singable melody with sparse accompaniment. A backing groove or written description alone is not a melody demo. Establish the opening hook within 0–3 seconds and leave room for sung phrases, breaths and the closing CTA. Present the actual demos, durations, phrase/section maps and recommendation. Refine within scope and wait for explicit approval of one exact audible demo/version. Do not write campaign hooks or sung lyrics before melody approval. Save the approved audio as the source for the song; a prompt or seed alone is insufficient.
2. Fit and approve the lyrics separately. Write opening-hook options and complete lyrics to the approved melody’s rhythm, phrase lengths, accents and pauses, using verified claims and the approved language and CTA. Check word count, syllables, natural stress, breathing room and intelligibility; simplify words rather than forcing pronunciation or rushing delivery. Map each line to the melody’s phrases and label future vocal timings as estimates. Present the full lyric specification and obtain explicit lyric approval before generating the sung track. Melody approval does not approve lyrics. If the tune must change, return to melody approval.
3. Generate and approve the final song separately. Use the approved melody demo as Source Audio in the installed ACE revision’s documented Cover / Remix workflow, together with the approved lyrics and vocal/arrangement direction. Inspect the supported inference interface, Source Audio guidance and UI support baseline, checking the required mode, model and source-adherence controls. Remove the instrumental-only setting for sung generation. A style-only Reference Audio input does not substitute for Source Audio. Source-guided generation cannot guarantee note-for-note transfer: listen to the actual result and compare the hook, melodic contour, rhythm, phrasing, pronunciation, breaths, structure and duration with the approved demo and lyrics. Report mismatches and return to the affected approval point for changes. Provide the actual playable final song for human vocal listening before transcription. Ask the user to listen to the complete performance and confirm intelligibility, pronunciation and the actual sung words against the approved lyrics; reuse specific listening feedback for that exact version. Automated transcription cannot replace this check. If listening is pending or words are unclear, do not finalize sung captions. After listening, transcribe the words actually heard, resolve mismatches at the affected approval point, and establish verified lyric timings. Present the measured duration, timestamped lyrics, rights status and limitations, and obtain explicit final-song approval. Lyric approval alone does not approve the generated performance.

Explicitly selected supplied/licensed track

Verify the selected recording’s source and permissions for the intended campaign, edits, placements and distribution. Do not assume a supplied file is licensed. Confirm the exact version, permitted excerpt, duration, vocal content where applicable, transitions and CTA space. Present the playable approved edit and verify beat/lyric timings as applicable. Do not replace the selected track with generated music without agreement.

Silent edit

Do not generate audio. Present a timestamped silent timing plan covering the opening hook, scene durations, reading time for captions and specifications, transitions and closing CTA. Obtain explicit approval of this timing plan before the test scene.

Approval point — final timing: Before Step 5, measure the actual approved final audio and establish a verified timing map for the applicable narration, beats, sections and sung lyric lines, or approve the silent timing plan. For sung music, complete the human vocal listening check before finalizing the transcript and timing map. Record the listener confirmation, audio version and timing-verification method; estimated or unverified timings must be labeled and resolved before synchronized visual production. Leave room for a clear opening hook and readable closing CTA. Do not cut off words or time scenes from estimated reading speed alone. Record the exact approved files/versions, rights evidence, production settings and approvals. If audio changes later, revise and reapprove the affected audio and timing before continuing.

5. Test one scene before full production

Create a visual reference sheet for the approved direction: product references, environment, character appearance where applicable, lighting, typography, palette and framing. Reuse the visual style selected in Step 3 and adapt its starter prompt to the approved brief and actual tool capabilities. Carry those references through every scene.

Generate or assemble one representative test scene in the target aspect ratio, selecting the hardest product-sensitive shot rather than the easiest attractive frame; for a non-product video, choose the most demanding scene for the selected format. Use the real product references and approved audio or silent timing plan. Show the user its frame and, when animation or product interaction is planned, a short moving test. Assess product geometry, packaging, logos and behavior independently of aesthetic appeal. Check product accuracy, readable labels, believable interactions, composition, cropping and space for captions or interface overlays.

Keep the product and key information away from screen edges and likely platform overlays. Add precise specifications, captions, logos and CTA text in the edit rather than relying on generated footage to spell them correctly. Avoid competing layers of captions.

Approval point: Ask the user to approve or adjust the test before generating the remaining scenes. Revise the test when needed. Do not use an attractive frame to conceal inaccurate packaging, changed ports or unsupported product behavior.

For Musical Cartoon Ads, retain its full animatic approval before animation. Then follow its Step 7 moving pilot of the hardest product-sensitive shot and obtain explicit acceptance before the remaining animation, within separately authorized pilot and remaining-production scope. This is staged execution within the existing workflow, not another numbered validation gate or a second test to repeat. Reuse the accepted pilot in the final edit. A still scene-image test does not replace this moving pilot. Inspect each subsequent output before assembly: compare against the approved references for packaging, product geometry, character identity, hands, lighting and environment continuity. Reject distorted text, unintended objects, morphing and unsupported product interactions. Reuse approved references or starting frames where the tool supports them; record any continuity fix and regenerate only the affected assets within the agreed scope.

6. Assemble and review the first cut

Produce the remaining approved scenes, then assemble them with FFmpeg or an equivalent editing tool the user has agreed to use. Match cuts to the approved narration, song and verified lyric timings, instrumental beat map, or silent timing plan, as applicable. Maintain consistent product appearance, character references and environment across scenes.

Add legible titles, captions when needed for the placement, verified specification callouts and the closing CTA. Use the approved destination URL in delivery notes; an on-screen URL is not automatically a clickable link. Keep text clear of platform interface overlays. Do not stretch footage to change its aspect ratio or add aggressive motion that distracts from the product.

Review the edit at the intended viewing size: Check that the opening lands quickly, each scene has enough time to understand, cuts and transitions serve the story, and the CTA is readable for a useful hold. Verify spelling, caption line breaks, contrast and synchronization. For sung music, keep a single lyric-caption layer based on the verified transcript. For instrumental-only, omit lyric captions and retain only purposeful approved information text and the CTA; spoken narration follows its approved caption plan. Composite exact logos and text in the editor when generated lettering is unreliable. Check framing and current platform overlays on the target aspect ratio, including the opening frame and closing card. Listen for intelligible speech, balanced music, clipping, abrupt audio cuts and unwanted generated sound. Check ordinary speakers or headphones and muted playback; a silent video must communicate through its visuals and text. Recheck product appearance and every factual claim against sources.

Export a playable MP4 suitable for the selected placement, using H.264 video, yuv420p, constant 30 fps, faststart and AAC audio when audio is present as the delivery preset. Silent edits remain silent. Agree on and record any necessary exception to the preset. When animation was selected, deliver real animated footage; a timed animatic or pan/zoom over stills requires explicit approval as a different format and a clear delivery label. Verify the current destination requirements rather than promising universal acceptance. Check the actual exported resolution, aspect ratio, duration and audio; inspect the beginning, middle, end and transitions, and watch/listen through the complete first cut.

If export or playback fails, inspect existing clips, timeline, render logs and actual file/stream metadata before considering regeneration. Repair assembly, encoding, audio mapping or container/playback issues first. An export failure alone does not authorize regenerating approved clips or making new paid calls.

Approval point: Present a reviewable first cut and ask for focused feedback on the hook, pacing, clarity, product accuracy and CTA. Apply the agreed changes and recheck affected timing, captions and audio. Do not publish or upload the video without explicit authorization for that destination and version.

7. Deliver the video and improve the workflow

Verify the actual final export: After any revisions, inspect the exported file’s dimensions, aspect ratio, duration, frame rate, video/audio codecs and file size against the current destination requirements. Open that exact file in a player and watch/listen through the complete video, including every transition, caption and closing CTA; scene previews or a render-success message are insufficient. Check for missing frames, black gaps, audio drift, unreadable text and export artifacts. Record the checked filename/version and actual review evidence, including pixel format and the required preset or approved exception. Record accepted Academy/Demo aesthetic limits separately from product accuracy; claim Production readiness only when all agreed checks pass. If full playback or audio inspection is unavailable, label that check unverified and give the exact next check; do not claim final readiness.

Deliver the final video file, a poster frame, caption text or subtitle file when used, the approved script and a concise source/asset list. Include editable project files or reproducible editing instructions when supported. Confirm the filename, duration, dimensions, aspect ratio, language and CTA destination. Separate completed files from items still requiring setup or manual work.

Provide a playable preview or accessible download of the actual video and supporting files. A storyboard, prompts, script, reference sheet or timed still animatic is not a finished generated video. Include a concise production summary: chosen approach, tools, approved assets, verification results, known limitations and remaining manual steps with their owner and next action. If production is blocked, name the exact failed or missing capability and offer the smallest useful action to continue; deliver completed materials honestly while keeping unfinished production explicit. Publishing or uploading requires authorization for the exact destination and final version, including any replacement action and companion copy. Reuse that authorization if already given; verify the supported action and its result before reporting publication.

For projects with audio, also deliver the approved final audio file and its verified timing map, plus editable narration/music components where supported and permitted. For newly generated sung music, include the approved melody demo, approved lyrics and timestamped transcript, final song, generation settings and rights/approval record. For instrumental or narrated projects, include the applicable music source/license and production settings. Keep local music-model and editing dependencies separate from campaign deliverables. Generate original music with ACE-Step through local Python; record the installed revision, model, prompts, settings and output files.

This studio is a workflow to test and refine. After the first attempt, ask the user:

1. Was it clear what to provide at the start?
2. Did the format recommendations fit the product and customer?
3. Which question, approval point or production step caused confusion or unnecessary work?
4. What would they change before trying another video?

Return a short workflow test recap: subject, chosen format, tools actually used, approval points completed, files delivered, blockers, user feedback and the next suggested adjustment. If the run stops early, record the last completed step and what is needed to resume. Keep that recap in the user’s chat or delivered files; this website has no feedback submission or project storage.

Offer one focused next iteration based on their feedback. Do not restart the whole process, rerun paid generation, or claim improved sales or conversions without evidence and authorization.