16:37
AI animation workflow
The video argues that a layered AI pipeline — Claude for scene-by-scene prompt writing, Soul Cinema for character/style keyframes, Stability 2.0 (referred to elsewhere in the transcript as 'Seedans') for animation and edits, and NanaBanana Pro for character fixes, with each prior scene's video fed into the next — lets one person produce a consistent multi-character, eight-style animated short film in a couple of hours, something animation 'used to' require studio budgets and years of training to do.
Framing claim: animation used to be studio-exclusive, requiring huge budgets and years of training; now it takes 'a couple hours and the right prompts.'
The short film has eight scenes, each in a completely different animation style/visual universe, generated live using Soul Cinema and Stability 2.0 on Higgsfield (the platform name is transcribed as 'Kickfield').
Story: a man wakes up, gets a stressed message from Anna, uses a watch offering a long route, a short route, and a mystery 'ultra fast' option to teleport across the eight worlds, ending up reunited with an equally disheveled Anna.
Character creation starts in Soul Cinema with a deliberately basic prompt ('cartoon highly stylized... a man waking up in the bedroom'); four generations cost only half a credit, so many batches are run to find the right look.
Soul Cinema's 'enhancer' toggle takes a bare prompt and expands it creatively into styles the creator might not have thought to request.
When a character looks off, the fix is to bring the image into Stability 2.0 and swap the specific element, not to regenerate from scratch ('that's the whole move').
Because the watch appears in every scene, a dedicated 'prop sheet' (material breakdown, internal parts, multiple angles) is generated via Claude + Stability 2.0 so the object stays consistent across all eight styles.
Per-scene workflow: upload the keyframe(s) and the previous scene's prompt to Claude, describe the new scene in a short beat list, and Claude outputs a full shot-by-shot Stability prompt.
Each scene is capped at 15 seconds, one generation per style, partly to make editing easier later.
Continuity across radically different styles is maintained by feeding the previous scene's rendered video (not just its prompt) into the next generation, along with the keyframe for style reference.
Specific prompt phrases act as high-leverage style/setting levers — swapping 'French graphic novel style' and 'futuristic environment' for other phrases would yield a completely different scene.
For the manga scene, a dedicated style-specific 'manga character sheet' is generated first (via Claude + Stability), separate from the original bedroom keyframe.
In the wedding scene, when the generated groom didn't match the established hero, NanaBanana Pro was used to swap the character to match scene one.
The last scene uses a stripped-down workflow — no Soul Cinema keyframe, no character sheet — just the watch prop sheet and the first video's prompt fed to Claude; Seedans reportedly invented Anna's full appearance from the phrase 'equally disheveled' alone while keeping the aesthetic consistent.
Closing pipeline summary: Claude writes the prompts, Seedans executes them, Soul Cinema and NanaBanana Pro supply the visual building blocks, and the previous video feeds into every new scene to keep the story consistent.
Promotional hook: Stability 2.0 just went fully global on Higgsfield with no waitlist and no business account required, plus unlimited access for 7 days, detailed further in the description.
Soul Cinema — The image-generation model the video calls 'easily the best model for this' for creating the protagonist's character keyframe and style anchors. Apply: Generate the character keyframe with a simple style-plus-scene prompt and run several batches (four generations cost about half a credit) to pick the strongest result before animating.
Enhancer (Soul Cinema feature) — A toggle in Soul Cinema that takes a minimal prompt and creatively expands it, surfacing stylistic directions the user might not have thought to request. Apply: Turn on the enhancer, type a bare prompt like 'cartoon style man waking up in the bedroom,' and let it generate varied stylistic interpretations to choose from.
Stability 2.0 ('Seedans') — The generation/editing tool used to animate keyframes into 15-second video scenes and to make targeted image edits or swaps; the transcript refers to it as both 'Stability 2.0' and 'Seedans.'. Apply: Feed it the scene's prompt, the style keyframe, and the previous scene's video to generate a consistent animated clip, or use it to edit one element of an existing image instead of regenerating from scratch.
NanaBanana Pro — An image tool used specifically to swap or fix a generated character so it matches the established protagonist. Apply: When a scene's generated character (e.g., the groom in the wedding scene) doesn't match the hero from scene one, bring the image into NanaBanana Pro and swap the character to align.
Claude prompt-scripting skill — Claude is used not to generate images or video but as the layer that turns uploaded reference images plus a plain-language scene description into a full shot-by-shot Stability prompt, via a dedicated 'Claude skill.'. Apply: Upload the relevant keyframe(s)/prop sheet and the previous scene's prompt to Claude, describe the new scene's beats in plain language, and use Claude's output as the ready-to-use generation prompt.
Prop sheet — A reference image generated for a recurring object (the watch) showing its material breakdown, internal parts, and multiple angles so it can be recreated consistently across every scene's art style. Apply: Upload the finalized character image to Claude, ask for a prop sheet for the object matching the exact style, generate it in Stability 2.0, and reuse it as a style reference in every later scene.
Character sheet (style-specific) — A dedicated reference sheet that redraws the protagonist in a new art style (e.g., manga, with ink outlines and screen-tone dots) before that style's scene is animated. Apply: Upload the original keyframe and the prop sheet to Claude, ask for a character sheet in the new style's terms, then generate it in Stability 2.0 before writing the scene's animation prompt.
Video-to-video continuity feeding — Feeding the actual previous scene's rendered video, not just its text prompt, into the next scene's generation so effects like the teleport portal, the character, and the style carry over automatically. Apply: When generating scene N, supply the keyframe for style reference, the prompt used for scene N-1, and the rendered video of scene N-1 so continuity is inherited rather than re-described from scratch.
Edit-don't-regenerate fix — Instead of re-rolling a whole new generation when a character or detail looks wrong, the flawed image is brought into Stability 2.0 and only the specific element is swapped. Apply: Take the off-target image into Stability 2.0 and issue a targeted edit instruction (e.g., 'swap the hair to black and put him in a suit') rather than restarting the generation.
15-second single-style scene cap — A production constraint where every scene is generated as one self-contained clip under 15 seconds in a single art style, ending on the teleport beat. Apply: Write each scene's prompt so setup, complication, and teleport-out resolve within a single 15-second generation, which simplifies later editing since each clip is a complete unit.
High-leverage style/setting phrases — The observation that a small number of specific prompt phrases (e.g., 'French graphic novel style,' 'futuristic environment') do most of the work in setting a scene's visual identity. Apply: Identify the style- and setting-defining phrases in a working prompt and swap only those to produce a substantially different scene while keeping the rest of the prompt structure intact.
Minimal-workflow finale (no keyframe or character sheet) — For the last scene, the creator skipped Soul Cinema keyframing and character-sheet generation entirely, feeding only the watch prop sheet and the first video's prompt into Claude to describe the ending. Apply: When the watch/prop and prior scene context are already strong continuity anchors, skip separate character generation and let the model infer appearance and consistency from prior-video context plus a text description alone.
The transcript's own terminology is inconsistent: the animation/video tool is called 'Stability 2.0' through most of the video but referred to as 'Seedans' at 12:19, 13:52, and 16:01, suggesting the two names point to the same product (consistent with the audience comment promoting 'Seedance 2.0') and that the video-generation model's actual name may have been mis-transcribed throughout.
The workflow solves three distinct consistency problems with three distinct mechanisms rather than one universal fix: a static prop sheet for the object that must look literally identical (the watch), a separate style-specific character sheet when the art style itself changes (manga), and video-to-video context feeding for motion/effects like the portal and teleport.
The 'two phrases' example (swapping 'French graphic novel style' and 'futuristic environment') implies that in this pipeline, prompt craft is less about prompt length and more about identifying the few phrases that actually steer style and setting.
The claim that Anna's full appearance was 'invented' from only the phrase 'equally disheveled' — with no character sheet or keyframe supplied — is presented as evidence that consistency can be inherited purely from prior-video context rather than from any static reference image.
The process keeps a deliberate human curation step at each stage (running batches and picking the best result, judging a look as 'off,' requesting a targeted edit like black hair plus a suit) rather than accepting the first generation, positioning the creator's editorial judgment as part of the 'workflow' being taught, not just the tools.
«Animation used to be studio-exclusive, requiring huge budgets and years of training. Now, it takes a couple hours and the right prompts.»
— 00:00
«So, Stability 2.0 just went fully global on Kickfield, and I mean everyone, anywhere, with no waitlist, no business account, just open. And to top it off, we're offering unlimited access for 7 days. More details in the description.»
— 00:10
«I built a short film. It's got eight scenes, but each one in a completely different animation style. Every scene is a different visual universe, and I'm generating all of it live using, hands down, the best combo right now. Soul Cinema and Stability 2.0 on Kickfield.»
— 00:27
«Now, to animate, I'm going to bring both images back to Claude and describe the scene.»
— 02:27
«Claude turns it into a full Stability prompt shot by shot.»
— 02:51
«All right, this is the part most people skip. And look, you probably get something decent with a basic prompt given how advanced Stability become, but if you spend the extra 2 minutes to build the prompt properly, the results are going to absolutely blow your mind.»
— 02:57
«Now, here's the workflow move that makes this whole thing work. I'm uploading that keyframe into Claude along with the prompt from the previous animation, so Claude knows exactly what happened in scene one. It keeps the style, the character, the teleport effect.»
— 04:26
«And then, I feed Stability the prompt, the keyframe for style reference, and most importantly, the previous video. That's how you keep a story consistent across scenes.»
— 04:58
«Soul Cinema gives you the style, but then Stability 2.0 fixes the character.»
— 06:14
«That's Stability understanding the visual language of manga. Not just aesthetic, but how the format actually works.»
— 09:54
«The demon reaction at the end, that sad clay pout, that's my favorite five seconds in the whole film. It's a reminder that the characters in these worlds actually react to him. They're not just background props. That demon had plans.»
— 11:08
«I never described what Anna looks like, not once. I just said she walks in equally disheveled. Seedans invented her completely and kept the whole aesthetic consistent.»
— 13:43
«Claude writes the prompts, Seedans executes it, and then Soul Cinema and NanaBanana Pro give you the visual building blocks. Oh, and the previous video feeds into every new scene, so the story stays consistent.»
— 15:58
Reception
Viewers admire the technical results but are frustrated by deceptive pricing claims, expensive credits, and discrepancies between the Claude skill shown and the prompts actually used in the video.
The video functions as a sponsored, replicable production tutorial: it walks through a concrete multi-tool pipeline (Claude for prompt-scripting, Soul Cinema for character keyframes, Stability 2.0/'Seedans' for animation and edits, NanaBanana Pro for character fixes) for keeping one character and one prop consistent across eight unrelated animation styles, framed inside an announcement of a limited-time free-access promotion for the underlying tools.
16:37