Skip to content

How do I write a good prompt?

2026-07-30 · help-centre article · Higgsfield · original

Topic: Prompting & AI assistants (also: Video generation & shot direction, Assets, references & consistency, Image generation & editing) · mode: Reference

This Higgsfield help-centre article on writing prompts says being specific helps a result more than anything else, because a short, vague prompt leaves the model to guess. It builds a prompt in layers: subject, details, environment and style, plus camera and mood for video. It then follows one character from a base shot through a restyle that says what to keep ("keep the exact same man, same face, same pose") to a video prompt written like a shot list. It ends with habits: reuse a character as a reference, write in English, and change one thing per try.

What is said

  • Need inspiration?

What is said

★ — how much it matters to the source.

  • ★★★ Higgsfield's help centre calls being specific the single biggest thing a user can do for a better result: a short, vague prompt leaves the model to guess, differently every time.

    “A short, vague prompt leaves the model to guess, and it will guess differently every time.”

  • ★★★ Higgsfield's help centre advises building a detailed prompt in layers: subject, details (appearance, clothing, expression), environment and style, and for video also camera and mood.

  • ★★ For a loose idea, Higgsfield's help centre suggests describing it in plain language to Supercomputer ("make a TikTok for my sneakers"), which fills in details, checks the plan, generates and routes each step to a suitable model. · Supercomputer, TikTok
  • ★★ Higgsfield's help centre says that, beyond loose ideas handed to Supercomputer, writing one's own prompt gives more control. · Supercomputer
  • ★★ To build a reusable base character, Higgsfield's help centre advises a clean, detailed description (age, hair, skin texture, exact clothing, background, lighting, even the camera type) in a neutral studio shot on a plain background.

  • ★★★ To restyle a character while keeping the same person, Higgsfield's help centre advises saying explicitly what to keep and what to change; the phrase "keep the exact same man, same face, same pose" holds identity steady.

    “keep the exact same man, same face, same pose”

  • ★★ To place a character in a real location, Higgsfield's help centre advises describing the setting in cinematic terms (lighting, depth, mood) so the character feels part of the world; that frame then starts the video.

  • ★★★ When animating an image, Higgsfield's help centre says the prompt should stop describing looks, which the picture already holds, and describe what happens: action, camera movement, timing and mood. · image-to-video
  • ★★★ Higgsfield's help centre likens a cinematic video prompt to a shot list, written in blocks: style and mood, cinematography and camera, lighting and color, beat-by-beat action, audio with dialogue timing, and a shot list with durations. · multi-shot prompt
  • ★★ Higgsfield's cowboy example is built as 6 shots over 15 seconds, each shot with its lens (35mm for the room, 75mm for faces), duration and action. · multi-shot prompt
  • ★★ Higgsfield's help centre advises matching the prompt's complexity to the result: a single line of motion works for a simple clip, and a short clip needs no six-block shot list.
  • ★★★ To keep a character the same across generations, Higgsfield's help centre advises generating it once and reusing it as a reference: Soul ID for a real person, an Element referenced in prompts for video, attached images in image models. · Soul ID, Elements
  • ★★ Higgsfield's help centre advises saying what is wanted rather than long "no this, no that" lists, though naming a few key things to avoid ("no readable text", "no plastic skin") can help on detailed shots.
  • ★★ Higgsfield's help centre advises making the prompt and the input image agree: a "still, calm" instruction fights a motion-blurred source frame. · image-to-video
  • ★★★ Higgsfield's help centre advises writing prompts in English, since models are trained mostly on English, with one exception: Chinese-developed models such as Seedance and Kling sometimes respond better to Chinese for nuanced motion. · Seedance, Kling
  • ★★★ To improve a disliked result, Higgsfield's help centre advises treating prompting as a conversation: generate, look, change one thing (the lighting, the lens, one wardrobe detail), generate again.
  • ★★ To build a longer sequence, Higgsfield's help centre advises using the last frame of one clip as the starting image for the next, then joining the clips in an editor. · keyframe
  • ★ For inspiration, Higgsfield's help centre suggests browsing generations in the Higgsfield Community and studying the prompts behind results close to the goal. · Higgsfield Community

Threads in this video

  • restate · Should prompts be short or long? — Higgsfield's help centre calls being specific the single biggest thing a user can do for a better result: a short, vague prompt leaves the model to guess, differently every time.
  • refine · Should prompts be short or long? — Higgsfield's help centre advises matching the prompt's complexity to the result: a single line of motion works for a simple clip, and a short clip needs no six-block shot list.
  • restate · Should you use an agent (Supercomputer, MCP) or drive the tools yourself? — For a loose idea, Higgsfield's help centre suggests describing it in plain language to Supercomputer ("make a TikTok for my sneakers"), which fills in details, checks the plan, generates and routes each step to a suitable model.
  • refine · Should you use an agent (Supercomputer, MCP) or drive the tools yourself? — Higgsfield's help centre says that, beyond loose ideas handed to Supercomputer, writing one's own prompt gives more control.
  • refine · How do you keep a character the same across shots? — To restyle a character while keeping the same person, Higgsfield's help centre advises saying explicitly what to keep and what to change; the phrase "keep the exact same man, same face, same pose" holds identity steady.
  • restate · How do you keep a character the same across shots? — To keep a character the same across generations, Higgsfield's help centre advises generating it once and reusing it as a reference: Soul ID for a real person, an Element referenced in prompts for video, attached images in image models.
  • refine · Should a video clip start from an image or from a text prompt? — When animating an image, Higgsfield's help centre says the prompt should stop describing looks, which the picture already holds, and describe what happens: action, camera movement, timing and mood.
  • refine · How do you direct a multi-shot sequence? — Higgsfield's help centre likens a cinematic video prompt to a shot list, written in blocks: style and mood, cinematography and camera, lighting and color, beat-by-beat action, audio with dialogue timing, and a shot list with durations.
  • first on record · Should a prompt list what to avoid? — Higgsfield's help centre advises saying what is wanted rather than long "no this, no that" lists, though naming a few key things to avoid ("no readable text", "no plastic skin") can help on detailed shots.
  • refine · What language and length limits apply to a video prompt? — Higgsfield's help centre advises writing prompts in English, since models are trained mostly on English, with one exception: Chinese-developed models such as Seedance and Kling sometimes respond better to Chinese for nuanced motion.
  • restate · When a generation fails, should you generate more takes, fix the prompt or start over? — To improve a disliked result, Higgsfield's help centre advises treating prompting as a conversation: generate, look, change one thing (the lighting, the lens, one wardrobe detail), generate again.
  • restate · How do you make a clip longer? — To build a longer sequence, Higgsfield's help centre advises using the last frame of one clip as the starting image for the next, then joining the clips in an editor.

Mentioned

Projects: Higgsfield Community

Organizations: TikTok

Concepts: image-to-video · keyframe · multi-shot prompt · negative prompt

Products: Elements · Kling · Seedance · Soul ID · Supercomputer

← Previous: What's in my Higgsfield account? · Next → 30-Day Seedance Unlimited on Higgsfield in 2026: What You Get and How It Works