Skip to content

How To Animate a Short Film with Blender + Higgsfield (Breakdown)

2026-09-23 · case-study tutorial · Amina · recording, 28:05 · watch on YouTube

Topic: Assets, references & consistency (also: Pre-production & story, Video generation & shot direction, Prompting & AI assistants, Failure modes & fixes) · mode: Tutorial · tags: one creator's experience, prompts shared, fiction, series, guest or collaboration

In the first Higgsfield Animation episode, animator Amina breaks down the car chase from Passport Rush, a five-minute action comedy by a traditional-animation team learning AI. The team works asset-first, making character, location and prop sheets in Soul 2.0, then uses grey Blender previews to guide movement and camera in complex shots, alongside very long, block-structured video prompts. Amina shares fixes for watercolor texture and camera angles, and says AI struggled most with subtle character acting, which the team calls the performance gap.

What is said

The parts of the video and what is said in each (★ — how much it matters to the video).

00:00:00 — Welcome to Higgsfield Animation: Passport Rush and a traditional-animation team learning AI

  • ★★ Amina opens Higgsfield Animation, a channel where Higgsfield's animation team shows its projects and breaks down its processes and pipelines as tutorials. — 00:00:00 · Amina
  • ★★★ Amina: Passport Rush, the animation team's upcoming short, is a five-minute action comedy about a girl who reaches the airport and realises she has left her passport across town. — 00:00:00
  • ★★ Amina says the Higgsfield animation team comes from traditional animation and had not worked with AI before joining Higgsfield. — 00:00:48 · Higgsfield AI

00:01:38 — Car chase preview from Passport Rush

00:02:23 — Pre-production: story, storyboards, a small team writing its own prompts, orange shots for Blender

  • ★★ Amina: The Passport Rush car sequence was made to test how 3D previs and AI could work together in an action scene. — 00:02:23 · previsualization
  • ★★ Amina: The Passport Rush team was small: people wore several hats, and with no separate prompt engineer, the character, environment and animation artists wrote the prompts for their own parts. — 00:03:03
  • ★★★ Amina: On the storyboards, shots marked orange get a Blender previs stage: complex shots with specific movement or camera angles, or anything that takes long to explain to the model in words; simpler shots go straight to video generation. — 00:03:30 · Blender, previsualization
  • ★★ Amina treats storyboards as a guide, not something fixed: since video generation is unpredictable, the team sticks to story beats and main scenes and leaves some shots to chance. — 00:03:30

  • ★★★ Amina: The team's pipeline rests on what it calls the asset-first approach: the quality of the image inputs dictates the quality of the film. — 00:03:30

    “We call it asset-first approach.”

00:04:09 — Art direction and the asset-first approach; the main character's sheet from a Claude prompt in Soul 2.0

  • ★★ Amina says unclear, sloppy or inconsistent image inputs make the whole film sloppy. — 00:04:21
  • ★★ Amina: The team created a role called asset managers: the people preparing character sheets, prop sheets and backgrounds make sure they all feel like one film and one world. — 00:04:21 · character sheet
  • ★★ Amina: For Passport Rush the team chose stylized 3D characters and watercolor backgrounds. — 00:04:50
  • ★★ Amina: To save time, the team generates initial character designs from prompts, then edits them by hand. — 00:05:19 · also in: Why I Switched to This 3-Day Animation Workflow
  • ★★ Amina: The team usually writes its prompts with Claude, though Amina says any other AI chatbot can be used. — 00:05:19 · Claude
  • ★★ Amina: The team checks Claude's character-sheet prompts before generating and corrects any detail that does not match its designs. — 00:05:46 · Claude, character sheet
  • ★★★ Amina: The team made all the assets of Passport Rush with Higgsfield's own image model Soul 2.0. — 00:06:12 · Soul 2.0, Higgsfield AI
  • ★★ Amina says Higgsfield's creative and engineering teams developed Soul 2.0, curated by people with backgrounds in film, animation, photography, fashion and graphic design. — 00:06:12 · Soul 2.0
  • ★★ Amina: Character sheet generation: open Cinema Studio on Higgsfield, pick Soul 2.0 in the model picker, paste the prompt, set the aspect ratio, choose a batch of four and generate. — 00:06:41 · Cinema Studio, Soul 2.0, character sheet

00:07:55 — Refining the character in Photoshop; designing the taxi driver

  • ★★ Amina: The team raised the brightness and saturation of the chosen character in Photoshop and painted strokes around her silhouette for a hybrid 3D and hand-drawn look. — 00:07:57 · Photoshop
  • ★★ Amina: Once the main character was locked, the team developed the other characters, such as the taxi driver, by the same method. — 00:08:27
  • ★★ Amina: Among the taxi driver's character-sheet candidates, one had a different hairstyle in every view and another had skin and facial hair too plasticky for the film's style. — 00:09:06 · character sheet, AI look

00:10:01 — Locations: watercolor tests and the late-afternoon lighting fix

  • ★★ Amina: Pure watercolor locations kept failing in tests, and the team seriously discussed dropping the watercolor look and pushing the film towards 3D. — 00:10:31
  • ★★★ Amina: Midday light made the Passport Rush locations flat and dull; switching to late-afternoon light through the prompt gave warm light and harsh shadows with more volume and depth. — 00:11:00
  • ★★★ Amina calls testing assets in video early in pre-production crucial, because visual problems can then still be fixed quickly. — 00:11:30

00:12:02 — Designing the taxi, the trucks and the explosion effects sheet

  • ★★ Amina: When the team already knows how something should move, animating it in Blender can be quicker than explaining every detail in a prompt, and it gives the model something to follow. — 00:12:28 · Blender
  • ★★ Amina: In video, the watercolor texture painted on the first taxi designs looked laid wrongly onto the 3D form, cheap, like a PNG stretched over 3D geometry. — 00:12:57
  • ★★★ Amina says heavy watercolor texture is a lot for a video model to hold once things move, and a watercolor prop on a watercolor background doubles the load. — 00:13:24 · text-to-video
  • ★★ Amina: Claude turns the short taxi request into a full prop-sheet prompt, which the team ran through Soul 2.0 on Higgsfield. — 00:13:24 · Claude, Soul 2.0
  • ★★ Amina: The fix for the taxi: the car became clean 3D and the paint stayed in the world around it. — 00:13:54
  • ★★ Amina: For a roofless version of the locked taxi the team edited the existing image in Seedream 5.0 Pro, which it uses for such edits, instead of generating from scratch. — 00:13:54 · Seedream 5.0
  • ★★★ Amina: The team builds asset sheets only for props that reappear from scene to scene and must stay consistent, such as three recurring trucks; the rest comes from text-to-video. — 00:14:23 · text-to-video
  • ★★ Amina: For the explosion effects sheet, asking for a mushroom-shaped explosion gave a literal mushroom of smoke; rewording it as "nuclear blast" more or less fixed it. — 00:14:59

00:15:30 — Production: animating a grey preview in Blender

  • ★★★ Amina: Most Passport Rush shots are generated as video from text with the assets attached as image inputs; some shots first get an extra preview stage. — 00:15:31 · text-to-video
  • ★★ Amina says the preview stage helps lock the camera angle of a shot. — 00:15:31 · previsualization
  • ★★★ Amina: The team exports the animated scene as a plain Blender preview: no colour, no materials, only the grey geometry, the motion and the camera. — 00:16:04 · Blender

00:16:22 — Video prompting: the blocks of the prompt and the Claude skill

  • ★★★ Amina: The team writes very long prompts even for a four-second shot, because every detail not written is decided by the model, which Amina calls a terrible guesser. — 00:16:34
  • ★★ Amina says the order of a video prompt matters: the model reads it from top to bottom, the first 20 (as heard) of the prompt matters most, and later text is followed less strictly. — 00:16:34
  • ★★★ Amina: The first block of the team's video prompt is scene context and style: what happens in the shot in plain sentences, plus the style rules and the physics of the world. — 00:17:03
  • ★★ Amina: The active references block lists the attached images or videos like a call sheet and says what each one is, or the model will guess. — 00:17:03
  • ★★ Amina: In the active references block the team describes the assets' qualities but adds no visual description, since the image inputs carry the look. — 00:17:33
  • ★★★ Amina recommends adding the line "Characters, environments, and props 100% match the reference" to a video prompt to keep the output true to the visual inputs. — 00:17:33
  • ★★★ Amina: The shot structure block of the video prompt writes down each shot divided by hard cuts and describes the action beat by beat with timing, like a storyboard panel in words. — 00:17:33
  • ★★ Amina: The locks and constraints block of the video prompt holds rules the model must follow strictly, because video models love adding their own ideas. — 00:18:07
  • ★★★ Amina: The team always puts "no music, SFX only" in its video prompt constraints, because that helps in post-production. — 00:18:07
  • ★★ Amina: The team's video prompts have a section asking for the 12 principles of animation (squash and stretch, anticipation, follow-through); the model does not follow all of them consistently, but they bring the movement closer. — 00:18:33
  • ★★ Amina: The team wrote its video-prompt structure down for Claude once as a skill, or system prompt, and applies it to every prompt. — 00:18:33 · Claude
  • ★★ Amina: The team shares the same system prompts so everyone's prompts follow a similar structure. — 00:19:03

00:19:09 — Generating the truck-squeeze scene from storyboard, Blender preview and references

  • ★★★ Amina: For the truck-squeeze shot the team attached the grey Blender preview in Higgsfield as the video input, then image references for the taxi, both trucks, the characters, the road and the lighting. — 00:20:01 · Higgsfield AI, Blender
  • ★★★ Amina: In the team's scene prompts, the video input guides the movement and camera while the image references show how everything should look. — 00:20:01
  • ★★ Amina: After generating a shot, the team watches it against the Blender preview, checking timing, camera and whether the car and characters still match the designs. — 00:20:30 · Blender

00:21:12 — Testing designs before production: three lessons

  • ★★ Amina: Lessons from the test stage between pre-production and production become the project's visual rules. — 00:21:33
  • ★★★ Amina: Early video tests showed how much watercolor texture to keep on the taxi; carrying that direction into the prompt kept surfaces consistent, with no photorealistic car in a painted scene. — 00:21:56
  • ★★ Amina: In tests run before the character sheets were ready, random people kept appearing in the empty taxi, though they were in no reference or prompt. — 00:21:56 · character sheet
  • ★★ Amina: Once the team added its characters to the taxi tests, the random people stopped appearing. — 00:22:25
  • ★★ Amina: A Blender preview with the camera too close showed mostly large grey shapes with little of the surroundings, so the video model could not tell how characters and objects were placed. — 00:22:53 · Blender

00:22:54 — Challenging shots: zooming out the preview, keeping camera angles and character details

  • ★★★ Amina's rule one: if the model does not understand a video reference, it probably cannot see enough of it, so zoom the camera out. — 00:23:32
  • ★★ Amina: Regenerating a tested seatbelt shot with the girl's final design, the same prompt and the other references, the camera kept changing angles between takes. — 00:23:32
  • ★★★ Amina: To hold the framing, the team took screenshots of takes with the wanted camera angle and attached them as references for the next attempts. — 00:24:10
  • ★★ Amina: The girl's headphones went missing in every new take because the screenshot used as a reference lacked them too. — 00:24:10
  • ★★★ Amina: The team drew the missing headphones back into the screenshot reference by hand in Photoshop; the next generation had them and kept the camera angle. — 00:24:40 · Photoshop
  • ★★ Amina: A gag inside the taxi, the girl ripping off the ceiling handles one after another, did not make the sequence because the team has not got it working yet. — 00:24:40

00:25:04 — Why AI struggles with subtle character acting: the ceiling-handle gag

  • ★★ Amina: Much of the ceiling-handle shot was keeping the handles consistent: the prompt had to say where they were attached, how they came off and how they looked in her hands. — 00:25:25
  • ★★★ Amina: The team could not get the girl's acting beats in the handle gag, a panicked grab and a pause of realisation, through text prompts alone, so it turned to Blender. — 00:25:54 · Blender
  • ★★★ Amina: With a simple grey character in the Blender preview, expressions and hand movements did not come through; with more detail, the model copied the Blender geometry instead of the character sheet. — 00:26:23 · Blender, character sheet

  • ★★★ Amina: The team calls it the performance gap: knowing how the character should react and pause, yet not getting that performance while keeping her final design consistent in the generated shot. — 00:26:23

    “We call this a performance gap.”

  • ★★★ Amina: The team expected the big action scenes to be harder, but Amina says AI struggles instead with subtle character acting. — 00:26:23

00:27:09 — Final lessons: 18 scenes in three days, the full breakdown to come

  • ★★ Amina: The team worked on the 18 scenes of the Passport Rush car sequence in three days. — 00:27:14
  • ★★★ Amina says that once Passport Rush is finished the team will share its full breakdown: production numbers, total generations, budget spent and how long the film took. — 00:27:14 · Amina

Threads in this video

Where the stories that moved here went next

Mentioned

Central: Amina · asset-first approach · Blender · character sheet · Claude · Higgsfield Animation · Passport Rush · performance gap · previsualization · Soul 2.0

Also: 12 principles of animation · AI look · Cinema Studio · Higgsfield Animation video prompt skill · Photoshop · product sheet · Seedream 5.0 · text-to-video

Also centred on Higgsfield Animation: Why I Switched to This 3-Day Animation Workflow

Replied to by

← Previous: Changelog: Production Skills Bundle · Next → Changelog: Higgsfield API cashback