Storyboards or Shot Lists? How We Plan Seedance 2.5 Shoots Now

For most of this year the answer to "how do I plan a Seedance shoot" was the same: build a storyboard first. Sketch it in GPT Image 2, lock the camera angle and the action panel by panel, then animate. We taught a whole module on that method. We retired storyboards as a reference in August, here is why. The honest answer is not that storyboards stopped working. It is that Seedance 2.5 changed what the bottleneck is.

What the storyboard method was solving

The original method, built for Seedance 2.0, used GPT Image 2 to sketch each shot as a rough, uncoloured panel, camera angle, composition and action locked in, with a short caption underneath describing what happens. Deliberately a sketch, not a photorealistic still, because a photorealistic storyboard panel drags its own colour grade and lighting into the final generation whether you want it there or not. The colour and grade came from a separate location or character reference instead. That separation was the whole point. The sketch told Seedance what to do, the reference told it what it should look like.

It worked, and it saved money, because you could see a shot was wrong on a cheap GPT Image 2 generation before spending on a Seedance render. For people who struggle to picture a shot in their head before it exists, that is still true today.

Why I took it out of my own process

My reasoning comes down to counting the places a project can go wrong. With just character references, a location reference and a written prompt, there is exactly one place a mistake can happen, the text. If the references are high quality, and you check the prompt Claude wrote for hallucinated details before you spend, the only failure left is Seedance itself misreading clean instructions.

Add a storyboard into that stack and you add a second place things can break. Now when a generation comes out wrong you cannot tell whether Seedance misread the prompt or misread the sketch. A storyboard is a layer more where Seedance can mess up, and it eats into your reference budget. Seedance 2.5 takes up to 50 references in one generation, and a storyboard panel was originally invented to compress a lot of information into a small number of slots. With 50 slots available, that compression is no longer the constraint it used to be.

What replaced it

The new method, from my own recent projects, is character references plus a location reference plus a script or shot list, with every piece of direction living in the prompt text itself rather than in a drawn panel. I now prefer a shot list over a storyboard, because a shot list is data, hard data I can upload into Claude, and Claude handles written technical instruction better than it handles interpreting an image. A shot list also holds up on a real project. My most recent piece, roughly two minutes long, ran without a single storyboard image, built entirely from a script and a shot list I had already worked out in my head before I ever opened Seedance.

What that shot list actually contains

While working on automating a storyboard-to-video pipeline, I build the shot list as a spreadsheet with a column for shot size, camera angle, movement, lens, the action happening, the audio, and how many seconds each shot runs. That spreadsheet gets uploaded into Claude alongside the references, and Claude writes prompts from it, shot by shot, with the right character and location tags already attached to each row. It is the same discipline a storyboard enforced, angle, action, timing all decided in advance, just carried in text instead of a sketch.

Where a storyboard still earns its place

This is not a blanket rule against sketching. The sketch storyboard method stays valid for Seedance 2.0, where the reference budget and generation length never demanded the same compression Seedance 2.5 solves differently. It also stays valid for anyone who genuinely cannot picture a shot without seeing it first, which is most people starting out. There's a case for animation specifically too. In 2D work a storyboard reference still performs well, because improvisation inside a locked sketch is part of what makes animation feel alive, unlike live action where it fights the realism you are chasing.

The trap either way

Whichever method you use, the failure mode is the same one I keep repeating. Claude invents content. It will write "clean-shaven" or "red shoes" or a cape into a prompt that never asked for any of it, because certain phrasing triggers associations in the model. Read every prompt line by line before you spend a generation on it. That habit, not the choice between a sketch and a spreadsheet, is what actually protects your budget.

Common questions

Should I still use storyboards for Seedance 2.5?

Not as the default. As of September 2026 the better-tested method is character and location references plus a written shot list, all detail carried in the prompt. Storyboards remain a fair choice if you cannot picture a shot without sketching it first, and they still work well in 2D animation.

Why did storyboards fall out of favour?

Because they add an extra layer where the generation can go wrong. With just references and a prompt, any mistake lives in the text, which you can read and fix. Add a storyboard and you can no longer be sure whether the sketch or the prompt caused a bad result.

What should a shot list include?

Shot size, camera angle, camera movement, lens, the action in the shot, the audio, and the length in seconds, laid out so it can be uploaded and read as data rather than interpreted as an image.

Is the sketch style from Seedance 2.0 storyboards still useful anywhere?

Yes, for Seedance 2.0 projects and for 2D animation, where a locked sketch still performs well. For live action work on Seedance 2.5, a shot list has replaced it in most workflows now.

Want the full method and the community that runs it every week? Join GenHQ.