How to read a recipe

Eighteen complete Seedance 2.5 prompts, one for each kind of shot you will meet most often. Each one is written for this guide, for an invented film and an invented brand. Copy the structure, and write your own film into it.

Every recipe is one generation, filled from a template, sealed and validated, and each has its own page. Read the short notes first: they say when the recipe fits and what to look for on the take.

The numbers follow the list below, and each page keeps its number. The four sections group the recipes by kind of shot, so a section's numbers are not consecutive: the picture-led section holds 1, 2, 3, 5, 6, 7, 10 and 11. Find a recipe by its number:

Templates A and B (one continuous shot; timed shots with written cuts) are taught in Video with Seedance → Templates A, B and D; C, E and F (the sound-off speaking plate, the audio-referenced shot, the singing plate) in Speaking shots and lip sync → Templates C, E and F. The risk bands are explained under Budget, below.

Words in the recipes

• A character sheet and its identity close-up (the role line names which panel gives the face and which the wardrobe): Photoreal anchors (Nano Banana Pro) → Character sheets. A state sheet shows one visible state (an outfit, a hairstyle) of the same character; a master plate and a reverse plate show a place from its two sides; a prop sheet shows one object.

• A plate means two things here. A master or reverse plate is an approved still of a place, used as a reference. A sound-off speaking plate or a singing plate is a generated video take, made for lip sync (Speaking shots and lip sync → Route 2, the Arabic default).

• The style anchor, the one accepted image that owns a drawn film's medium: Stylised films (Seedream 5.0 Pro) → The style anchor.

• Omni-reference is the platform mode that takes image and audio references (Video with Seedance → Platform facts and settings). Sealed means the prompt carries no titles, scene numbers or "as above", because the model remembers nothing (Video with Seedance → The prompt grammar); validated means it has passed the checks made before sending (Video with Seedance → Validate, then pre-flight).

• KEEP CONSISTENT lists, BACKGROUND lines and the count lock in multi-shot prompts: Video with Seedance → Multi-shot formats.

• A named exception, such as a locked camera the plan names for one shot: Camera and lens → Named locks and breathing frames.

What each recipe gives you

• When to use it: the shot it was written for, and the invented film it comes from.

• Methods: the methods the shot rests on, each with its label. An option is named first, then its trigger, then the reason this film chose it.

• Shared rules: a pointer back to this page, for the rules every recipe keeps and the notes on its section.

• References: the images in upload order. That order sets the @image numbers in the prompt.

• Settings: the parameters sent with the job: the mode, the length, the aspect ratio and the audio.

• Cost: the draft and the finish, in Higgsfield credits. "cr" means Higgsfield credits: a 480p draft costs 3 cr a second and a 1080p finish 16 cr a second, as checked on the platform in October 2026. Each cost shows its sum (length × rate). Check your prices before each take.

• Budget: the shot's risk band, with its expected spend and its ceiling, each shown as length × rate. A low-risk shot is budgeted at 31 cr a second expected and 62 at the ceiling. Low risk means a product, a place, or a figure whose face doesn't read. A high-risk shot is budgeted at 62 and 123. High risk means any of: a readable face, hands or touch points, product physics, more than one acting character, 3–4 shots in one clip, an end frame, a new method, or a hero speaking shot (its lip-sync pass included). See Planning, budgets and takes → Costs and the estimate.

• What to change, what to keep: the template the recipe follows, the lines that are yours to rewrite for your film, and the lines that stay word for word (see What to change, what to keep, on this page).

• The prompt: the full text, already validated. Each window ends where the next begins, the last ends at the duration, the AUDIO times match the timeline, and every tag exists and appears only where its subject is visible.

• Check the take and the common failure and its fix: what to look at first on this kind of shot. The full list is in Video with Seedance → The failure table.

• Sources: the creators and films behind the recipe's methods.

What every recipe keeps

• The look line sits late, in STYLE, and carries no grain. Grain is added once, in finishing (see The grade and texture → Texture once). Each film writes its own look line once and pastes it unchanged into each of its prompts.

• Options are chosen per film, in the plan. Some options say "when the plan names it": that means you decide per film, in your shot plan (see Planning, budgets and takes), and the topic that teaches the option says what favours it. Each recipe names the option, gives its trigger, then says why its film chose it. In these recipes: the delivery map (recipes 8 and 14), the hero colour leading (recipes 1, 14, 15, 17 and 18), one locked lighting formula (recipe 18), the smiling close-up (recipe 14) and Route 1 for an Arabic hero line (recipe 12).

• These rules hold in every prompt (decided):

  1. The client's product is always pristine, with no wear or sign of use.

  2. No storyboard image is attached, in any role.

  3. The motion standard: every shot stunning and dynamic next to the treatment, the anchors and the plan, with camera, acting, body, hair, hands, face, the product's texture and light moving together; each shot distinct in acting and camera, yet fitting the rest. A slow or static camera where the plan wants motion fails, whatever else holds (see Video with Seedance → Take selection and review).

  4. A named lock (recipes 5 and 6) is judged as a lock, with the talent and the place still alive (see Camera and lens → Named locks and breathing frames).

  5. The camera during speech: during the words the camera breathes or pushes very slowly, with bigger moves before or after the line (see Speaking shots and lip sync → The camera during speech).

  6. Live motions: the place keeps the live motions your plan names: 2–4 outdoors or wherever the place has them, as in a busy café; 0–1 in a quiet room (see Video with Seedance → The fixed set and the live place).

• The generation rhythm (decided): one shot per scene is made first as a sample shot, in two 480p drafts, and all the samples are reviewed together; each remaining shot then gets one new take in each batch (see Planning, budgets and takes → The generation rhythm). A failed take is never fixed at the pixel level, by hand or with a painting or retouching tool. It is regenerated with one diagnosed change, or given one model edit pass (see Planning, budgets and takes → Fixing a failed take).

Picture-led recipes

Eight shots where the picture carries the film and nobody speaks a principal line: recipes 1, 2, 3, 5, 6, 7, 10 and 11. They use Templates A and B (see Video with Seedance → Templates A, B and D).

Ad-format recipes

Three short-form formats where the ad starts and ends on the product: recipes 15, 16 and 17. Their biggest risk is a catalogue feel, so each one carries an acting task or a chain of cause and effect.

Stylised recipes

Two drawn films: recipes 4 and 18. Both start from one accepted style anchor, made in Seedream 5.0 Pro (decided), attached once per generation. Describe only what changes, because the model pushes drawn work toward volume (see Stylised films (Seedream 5.0 Pro) → Seedream for stylised films).

Speaking and singing recipes

Five shots where a face says or sings a line: recipes 8, 9, 12, 13 and 14. The same rules run through all five:

• Templates and route: they use Templates C, E and F (see Speaking shots and lip sync → Templates C, E and F), and the route for each is chosen in the plan (see Speaking shots and lip sync → Choosing the route per shot).

• The camera during speech: during the words the camera breathes or pushes very slowly, the push written large enough to show; bigger moves come before or after the line (decided). As a working figure [best available, not yet validated], the face grows by about a fifth over the line, from a push of about a sixth of the camera's distance (20 cm at 1.2 m, 35 cm at 2 m).

• The angle: the face stays within about 45° of the lens.

• The native reviewer: every Arabic line goes to your native reviewer before use, and Saudi and Gulf lines stay drafts until they pass (see Voice and Arabic dialects → The native reviewer).

• The placed voice: each on-screen line is placed in its shot's space in the mix, with the dry recording kept muted beneath it (see Sound design and mix → The placed voice).

What to change, what to keep

A recipe's product, people and place belong to its invented film. Its protective wording belongs to every film. Each recipe names the template it follows, the lines that are yours to rewrite, and its own lines to keep. Most of the kept wording is the same in every recipe.

Yours to rewrite:

• SCENE CONTEXT;

• each role line's subject (the alias, age, build, wardrobe and what is in each hand);

• LOCATION MAP's layers, counts and live motions;

• FIRST FRAME AND BLOCKING;

• the field of view in OPTICS;

• the CAMERA rig, height, distance, move and its degrees;

• the times in ACTION / TIMELINE and AUDIO;

• PERFORMANCE;

• the weights in PHYSICS;

• the materials in TEXTURE;

• the source, Kelvin and ratio in LIGHTING;

• the counts in POSITIVE LOCKS;

• the look line in STYLE (your film's own).

Kept word for word:

• each role line's scoping clauses: "This sheet gives identity and wardrobe and nothing else …", "used for the face alone", "as architecture and light only", "the set holds exactly as pictured", "it does not decide where the camera stands", "where these words and the picture differ, follow the picture" and "it is not a start frame";

• "(in motion from frame one)" in the FIRST FRAME AND BLOCKING header;

• the FORMAT MODE line, and "the lens holds steady" in OPTICS;

• the cut sentence that closes CAMERA: "The cut comes after the beat …" or "The cut comes in motion …";

• "No surface lights itself." in LIGHTING;

• the closing sentences of a sound-effects AUDIO block: "No music and no voices. Nothing rises, swooshes, stings or fades at the opening, at a cut, or as the clip closes. No subtitles and no on-screen text.";

• the end of STYLE, after your look line: "light seen only where it lands, as highlights and reflections on real surfaces";

• the forms in POSITIVE LOCKS: "Exactly …" for every count, and one "The take fails if …" sentence.

Sources

• Higgsfield's Seedance 2.5 prompting guide and prompt builder (the block order, the anchor line, the late style line, the degree ladder), behind every recipe.