The concept calls for generated imagery. First question: which visual treatment
makes this idea legible in this brand's tone? Use refs/brand.md
(guide-capture-a-brand) and the approved concept, not a preferred style by habit.
Show a few relevant style frames, then lock the pick. The concept itself is
llm-pick-a-visual-metaphor's; the prompt grammar is nano-banana-prompt-format.
The table is a set of starting treatments with known craft. It does not require a character or a miniature world. A composition made from existing imagery or designed graphics follows its own workflow and does not need this generated-stills process.
Routes
| Look | Pick it when | Resolution and cost | What you get | Load |
|---|---|---|---|---|
| Abstract forms or materials | A transformation, pattern or physical relationship carries the idea without a character | Depends on fine detail and motion; estimate after the test design | Objects, textures or forms that express the concept in the brand's tone | nano-banana-prompt-format, guide-choose-a-video-model |
| Claymation | A warm, handmade toy world; ads and news shorts that should read as clearly made, not motion graphics | H3 Max 768P, about $0.09/s | A tilt-shift miniature; clay animates cleanly | style-claymation |
| 3D animation | Expressive characters or dimensional objects suit the concept and brand | H3 Max 768P for the worked character treatment | A dimensional animated look; the character recipe is one treatment | style-3d-animation |
| Paper-cut diorama | A gentle story short; a post or a lesson retold | H3 Max 1080P (twice the 768P price): the fibre is the look | Layered paper in one set | references/looks.md |
| Paper collage | Editorial, a news short | H3 Max 768P | Printed photos cut out, ransom-note type | references/looks.md |
| Ink comic book | Energy, music, a punchy social ad | H3 Max 768P | Ink, halftone, misregistered colour | references/looks.md |
| Two-colour risograph | A graphic print look | Untested past the style frame | Two inks and grain | references/looks.md |
| Drawn 2D cartoon, or a mascot's own style | A brand mascot; a kids' video | H3 Max (or Kling for baked lettering) | Flat colour and line, on model | references/looks.md |
| Photographic | Real-looking materials, objects or scenes serve the concept, alone or in a mixed treatment | Model depends on motion and detail | Photographic imagery; illustrative scenes remain distinguishable from real evidence | references/looks.md, guide-choose-a-video-model |
Stills cost cents (nano banana 2.1 about $0.07 at 2K; Pro edit about $0.17); the animation is most of the spend.
Process
- Style frames. Select candidates from the brand direction and the concept's
needs; explain the fit of each. If the person supplied a treatment, develop it
rather than reopen the choice. Two to four frames of the same shot (the opening, or the one that
matters most) on nano banana 2.1, 2K, the deliverable's aspect: identical scene text,
only the style paragraph changed, looks that differ in kind (collage, risograph,
clay was a good spread). When the video is for someone specific, look at what they
already make: a motion-graphics audience is most impressed by a look motion
graphics can't do. Gate:
review_chooseof the frames. - Lock the style line. The picked frame's style paragraph goes verbatim into every later prompt, and into the job's notes. Fix the frame's own flaws now, in shot 1 ("pull the camera back so the ENTIRE fits").
- Cast when the concept needs it. Every generated character that recurs gets a sheet in this look
(
nano-banana-character-sheet), cut into views. Gate: a pick per character. An abstract, object-led or graphic concept may have no cast; use consistent object references and the style frame instead. - Stills, all at once, on nano banana 2.1 edit (Pro edit for big lighting changes)
with the picked frame as Image 1: "use it ONLY for its art style, materials,
lighting and palette. Do not copy its ." Then the style line, the scene with
the camera and which way any figures face, character views when needed, and, for
generated metaphor stills, "No real faces, no text, no logos"
(
llm-pick-a-visual-metaphor). Product-character shots preserve the packshot's exact label throughguide-keep-small-details; lettering belongs to that fidelity task. A recurring prop gets its earlier still as a reference too. A shot that is the same place in another state (lights off, after the fall) edits the earlier still, listing what stays; big lighting changes need Pro edit, as a numbered list. Plan multi-aspect framing now (guide-pace-a-short-for-social). Sheet every still yourself (python-contact-sheet,images) and zoom into papers, screens and devices for baked words and real logos before the person sees them; remove a baked word with a one-line edit (nano-banana-targeted-fix), "remove the word …, change nothing else". Gate:review_chooseper shot, all in the same turn. - One motion test, then the batch. Select a model for the shot's hardest demand
(
guide-animate-a-shot,guide-choose-a-video-model); the table's models are starting points. Quote the price, test a representative demanding shot with the look's phrasing, and Gate: a yes on its motion; then the rest at once. For H3, useh3-max-image-to-videoandh3-prompt-format. Every prompt: mouths closed and no speaking verbs, unless a line is lip-synced; big, counted motion; and lock in words what must not change from the first frame to the last (the sky's colour, a building left dark). Check a 4-frame sheet per clip (0, 33, 66, ~98%: seek no later than duration − 0.15 s or the frame is empty) and re-render only what breaks the story.
Choosing
- Text-free references. A style frame with lettering in it gets its lettering copied into new shots, partly, and adds sound words (HISS, SLAP!): make a text-free copy of the picked frame before using it as a reference.
- A busy style frame as the only reference: about half the stills copied its set. Describe each new location strongly and drop the takes that copy it before the review.
- Describe the technique, never a film, studio or artist.
- Spell out each person's defining features in every shot prompt, not only in the references: a two-person comic shot lost the second person's facial hair and age. Two people drifting towards each other can end up reading as romantic; give friends a beat that keeps space (a big high-five, both turning to camera).
- Resolution follows the look (Routes): pay for 1080P only where the texture is the look.
- Real people and places become metaphors, or follow
guide-depict-real-people-and-places. - Mixing looks in one piece: change the look at a story turn (the real world, then the product's world), not on a timer. A sheet per person per look keeps the cast recognizable across them. A free transition between looks: the character's front close-up from each look's sheet, swapped every 0.25 s over a blurred copy of itself.
Avoid
- A look described in words and never shown before the stills.
- Unlocked video prompts: H3 turned an overcast sky blue with cartoon clouds and lit up a building the story says stands empty.
- Big lighting changes on nano banana 2.1 edit: asked twice to switch a room's lights off, it kept them glowing.
- A stills rejection with no note answered by re-rolling everything: find the shot whose idea doesn't read, and retake that first.
- An unplanned change of treatment that breaks the piece's visual continuity.
Related
llm-pick-a-visual-metaphor: what each shot shows.guide-animate-a-shot,guide-choose-a-video-model: the video routes.guide-keep-small-details: a product's label or a logo kept exact.guide-design-a-set: one location shared by many shots.
