ChatGPT → Leonardo: The Exact Seedance 2.5 Workflow Behind The Match
Share
ARTICLE 4 · FIND YOUR VOICE → GPT → LEONARDO → QA → CAPCUT
The exact ChatGPT → Leonardo Seedance 2.5 workflow behind The Match
The lesson is not one magic prompt. It is the development path: human decisions first, ChatGPT turns those decisions into production instructions, Leonardo executes them, QA diagnoses the right layer, and CapCut packages the approved visual for delivery. The visual was then locked before the final sound layer was added with Suno using my own voice.
The jobs are different
Creator: decides meaning, audience, message, emotional goal, brand rules and what must remain human.
ChatGPT: structures approved decisions, identifies missing decisions, builds reference instructions and translates the creative brief into a platform-specific prompt.
Leonardo: renders the visual from the supplied reference, model settings and execution prompt.
CapCut Pro: finishes and packages the approved visual with precise editorial decisions the generation model should not control.
Suno: adds the final sound layer using my own voice as the human identity anchor while the picture remains locked.
Do not ask a production tool to solve a Voice problem that should have been decided before generation.
1 · Start with the Find Your Voice brief
- What should this make people feel?
- What should they understand or remember?
- What makes this recognizably mine?
FEEL: quiet recognition, warmth, creative connection and the moment something finally comes alive.
REMEMBER: creators can generate many things; the breakthrough is when something begins to feel like yours.
MINE: Bee Righteous + Match, black-and-gold visual language, restrained emotional storytelling, and “Create What You Love. Love What You Create.”
2 · Give ChatGPT an instruction framework before asking for a prompt
Work only from the creator decisions I provide. Preserve approved meaning, audience, message, tone, character rules and brand constraints. Do not invent missing identity decisions. If something important is missing, mark it [CREATOR DECISION REQUIRED] and ask one concrete question. Separate creative decisions from technical execution. When the creative layer is approved, translate it into a platform-specific production prompt without changing the underlying intent.
3 · Build the production reference before the video prompt
The first character bible was useful for human review but too busy as a production reference. It contained multiple poses, expressions, labels, diagrams and repeated versions. The fix was a single clean square reference: Bee Righteous + Match only, full body, three-quarter front, same baseline, neutral expressions, warm neutral background, no props, alternate poses, environment, readable text or extra logos.
Production principle: a presentation board and a model reference are not the same asset. The production reference should remove ambiguity.
4 · Leonardo setup used for refinement
| SETTING | VALUE | WHY |
|---|---|---|
| Platform | Leonardo AI | Visual generation environment. |
| Video model | Seedance 2.5 | Model used for the 15-second short. |
| Duration | 15 seconds | Five clear emotional beats. |
| Aspect ratio | 9:16 vertical | Short-form vertical delivery. |
| Refinement resolution | HD 720×1280 | Used while refining creative control. |
| Audio | ON | Tested ambience and timing with picture. |
| Private Mode | ON | Kept working generations private. |
| Legacy | OFF | Used the current generation path. |
Leonardo also offered Standard 496×864 and Full HD 1080×1920. Document what you actually deliver rather than rewriting history around an ideal setting: the approved production copy in this case study is 720×1280.
5 · Translate the Voice brief into a platform prompt
The production prompt was ordered by priority: character lock → emotional target → timed beats → camera → motion restraint → consistency restrictions → audio → final-delivery priorities. Longer is not automatically better.
6 · RUN 1 — FIRST DRY RUN
The core concept survived. Bee was readable, Match had charm and the reveal worked. The problems were concentrated in execution: the final third became too busy, generic music imagery weakened the world, spatial clarity dropped, the ending drifted toward a fantasy transformation and fingertip contact was not deliberate enough.
Do not reject the entire creative idea because one production layer drifts. Preserve what worked and diagnose the failed layer.
7 · RUN 2 — CONTROL THE DRIFT
The biggest improvement came from a cleaner canonical reference and tighter sequencing, not simply more prompt language. Character consistency, Bee/Match differentiation, recognition and the ending improved. Remaining issues were a too-static / too-sad opening, Match appearing too early, unclear fingertip causality, insufficient final hero hold and overly self-referential bee drawings on the desk.
8 · RUN 3 — PRODUCTION RUN
Bee thoughtful and dissatisfied, not sad.
Match remains hidden until Bee physically reveals Match. No pop-in.
Mutual eye contact, tiny mirrored head tilt, flame brightens, Bee smiles.
Both fingertips stay visible; touch is held; gold pulse begins exactly at contact.
The same room warms subtly.
Stable composition with dark negative space and minimal movement.
Platform-optimized production prompt
15s, 9:16 premium stylized animated short. Use the attached image as a strict identity reference; never change character faces, clothes or proportions. Bee is always larger; Match is smaller with the glowing flame chest mark. One continuous intimate nighttime creator studio. Warm amber/charcoal, matte cinematic finish, restrained adult-creator tone.
0–4: Bee at a dark wooden workbench with headphones, compact keyboard, notebook, abstract album-art/storyboard/waveform cards. Thoughtful and dissatisfied, not sad. Slow push. Bee moves one card aside.
4–7: Match is completely hidden before this. Faint gold glow appears behind the final object; Bee notices with eye/antenna reaction, physically moves the object, revealing Match curled asleep. No pop-in.
7–10: Match wakes. Hold a clean two-shot with direct mutual eye contact. Match gives a tiny head tilt, Bee mirrors, flame brightens, Bee smiles. No room transformation yet.
10–12: Match gently floats closer; Bee extends a hand. Both fingertips remain clear in the same frame; touch and hold briefly. Exactly at contact, one small gold pulse originates between fingertips.
12–13.5: The pulse subtly warms the same studio: practical lights, keyboard edge light, one or two art cards, a few particles. No scene replacement.
13.5–15: Stable hero composition hold: Bee larger, Match beside him, both readable, workspace behind, clean dark negative space upper third. Only tiny breathing/blinking.
Camera: slow, controlled, continuous-feeling; no rapid cuts or dramatic zooms.
Audio: minimal ambience, near silence at eye contact, one delicate chime exactly at touch, warm music bloom after. No dialogue or lyrics.
Do not generate: text, logos, UI, portals, new locations, floating music notes, explosions, duplicate/morphed characters, extra limbs/wings/antennae, or bee drawings on the desk.
Priority: reference consistency → hidden reveal → eye-contact hold → fingertip contact → touch-causes-transformation → final hero hold.
APPROVED FINAL VISUAL COPY
The visual production run is now locked.
The delivered case-study copy is 15.79 seconds · 720×1280 · 30 fps, with stereo audio in the source file. The CapCut package includes the visible production-run treatment, including “THE MATCH — PRODUCTION RUN 01” and “BRINGING MATCH TO LIFE.”
This file became the visual source of truth for the final Own Voice sound pass. Character and story decisions stayed locked while the audio layer was developed.
9 · Diagnose the correct layer before changing the project
The idea/message/Voice is wrong. Return to Flame / Voice.
The reference/context is ambiguous. Fix the source material.
The brief is right but the platform misses it. Change the platform-specific prompt, sequencing or settings.
For The Match, the core creative idea survived. The meaningful problems were primarily input + execution problems.
10 · Delivery QA measures Voice, not only visual quality
- Does it still communicate creative recognition?
- Do Bee and Match remain visually distinct and consistent?
- Is Match genuinely discovered?
- Does recognition have time to register?
- Is fingertip contact the readable cause of the awakening?
- Does the final frame support the Jack Righteous message and controlled typography?
11 · Controlled finishing: CapCut Pro
CapCut is no longer a future plan in this case study. It has been used to package the approved production visual. The important lesson is that generation is not delivery: the creator still controls what is labeled, timed, emphasized and exported.
Affiliate disclosure: I use CapCut Pro as part of this finishing workflow. I’m a CapCut affiliate, so I may earn a commission if you use my partner link. It does not change what I recommend or what I show in this Dry Run.
12 · The three-Short public case study
Purpose: show what worked and what drifted.
YOUTUBE LINK TO BE ADDED
Purpose: show reference + instruction improvements.
YOUTUBE LINK TO BE ADDED
Purpose: show the approved visual carried through the final Suno Own Voice sound pass.
13 · Add Your Sound completed the 3/3 production
APPROVED VISUAL → SUNO / MY OWN VOICE → SOUND QA → CAPCUT SYNC → FINISHED AUDIOVISUAL MASTER.
The dedicated sound article preserves the deeper Suno Own Voice workflow and the role of the locked visual as the source of truth.
Continue: Add Your Sound — Suno With My Own Voice →
14 · Save the development record for V10
- creator Voice brief
- GPT instruction version
- reference-image version
- Leonardo model / dimensions / settings
- execution prompt versions
- what worked / drifted / changed
- approved visual production copy
- CapCut finishing decisions
- Suno voice and sound record
- final audiovisual export
- published YouTube URLs as each Short goes live
Where CONTROL / Creator Pro begins
The free Dry Run gives the usable route. CONTROL / Creator Pro goes deeper into reusable GPT instructions, reference architecture, sound briefs, context preservation, controlled iteration, drift diagnosis, cross-tool continuity, version control and delivery standards.
