ComfyUI adds 9-panel storyboard handoff into Seedance 2.0 Reference-to-Image

• PurzBeats' ComfyUI workflow turns a single image plus a style reference into a nine-panel storyboard with Nano Banana 2, then hands those frames to Seedance 2.0's reference-driven generation.
• AIwithSynthia's match-day demo shows the same pattern in a different wrapper: make the storyboard first, then use that board as the visual reference for a multi-scene vlog in Seedance.
• PJaccetturo's Nexus breakdown and techhalla's Magnific tutorial both treat reference images as the real control layer, whether that is a character sheet, a location plate, or a single hero frame.
• PJaccetturo's trailer guide breaks Seedance prompting into concrete shot grammar, including geography, lensing, sound design, and cut-by-cut timing, instead of one giant mood prompt.
• PJaccetturo on 720p and finishing passes and dustinhollywood's extension claim both frame Seedance clips as short blocks, usually around 15 seconds, then stitched into longer edits elsewhere.

You can browse Leonardo's Seedance surface, open techhalla's Magnific Space, and PJaccetturo also linked a Luma workshop page while posting the underlying shot logic. The interesting bit is how quickly creators have converged on the same harness: storyboard first, ordered frames second, motion last.

The new wrinkle is not "use reference images." It is the handoff format. PurzBeats spells it out as a four-step chain:

1. Start with your source image plus a style reference.
2. Use Nano Banana 2 to generate a nine-panel storyboard.
3. Send those panels into Seedance 2.0 as the visual reference.
4. Write the motion prompt as a transition path between shots, using shot markers like [shot1] [shot2].

AIwithSynthia's match-day demo lands on the same idea from the other side. Instead of asking Seedance to invent a whole football vlog from scratch, the prompt tells it to use an uploaded storyboard image as the reference for character appearance, outfit, and scene progression.

The prompts getting traction are basically pre-vis docs in plain text. the match-day prompt maps a fan vlog as a scene list, while the runway prompt turns a fashion pageant into a nine-panel production board with a camera path.

The recurring structure looks like this:

• Reference rule: tell the model what image or board is the source of truth.
• Scene order: wake-up, prep, travel, arrival, finale, or any other fixed sequence.
• Camera behavior: handheld vlog, FPV path, tracking shot, pull-back.
• Continuity anchors: outfit, character identity, props, lighting style.
• Ending beat: a specific payoff shot, not a vague ending.

the runway board is especially literal about camera choreography. It lists eight contestants, routes the viewer through them with a red path, and even names the motion loop as "Walk In → Ramp Walk → Close-Up → Say Name → Pull Back → Next Contestant." That is storyboard language, not prompt-poetry.

Two creators pushed the same control trick in different directions. PJaccetturo says some Nexus sequences only needed five generated shots because Dreamina Seedance 2.0 could build coverage around a small set of "anchor images." techhalla goes even harder and starts a three-minute music video from one uploaded portrait.

PJaccetturo's earlier trailer guide adds the missing production logic. His workflow stacks:

• a start frame,
• a location shot,
• a character sheet,
• plus strict scene context about where each subject stands and moves.

That is close to how CuriousRefuge's character consistency test describes the process too: establish the wardrobe, environment, and visual language upfront, then the single character board carries consistency across multiple shots.

The most useful phrase in the evidence pile is probably "burst coverage." PJaccetturo says he borrowed the method from other creators, then used Seedance to generate 15 to 30 angle changes over a 15-second clip while asking the actors not to move much. The goal is not one perfect shot. It is a dense basket of cuttable options.

He describes two modes:

• Quick coverage: upload character sheets plus a location, accept looser blocking.
• Consistent blocking: use the 15-second, 15 to 30-angle burst and extract frames later.

pzf_ai's London sequence shows the lighter-weight version working fine for quieter material. One reference character moves through a cafe, riverside, taxi, and apartment without dialogue, which makes the workflow look useful for mood reels and brand films, not just action scenes.

One production constraint kept surfacing after the flashy demos. PJaccetturo says "15 seconds is best," runs most explorations at 720p, and only rerenders selected closeups in 1080p because those pulls are expensive. He also says many longer scenes are really collections of short generated segments.

dustinhollywood makes the sharper claim that Seedance has no true extension tool, and that anything longer than 15 seconds becomes either an in-window extension trick or a last-frame handoff into a new clip. That lines up with the broader editing behavior in the creator threads: Seedance is being used like a shot engine, then stitched, upscaled, or reframed in tools like Leonardo, Magnific, Luma, or Topaz.

That short-block assumption also explains why awesome_visuals' Topaz post is about cleanup, why DrSadek_'s short film teaser is a 15-second insert, and why the storyboard-first setup is spreading so fast. Once the model only has to solve one shot block at a time, the board does the continuity work.