Higgsfield: how to write prompts the model actually understands
Higgsfield · Updated:
Higgsfield is a platform with proprietary models — Soul 2.0 (image, up to 4K), Soul Cinema (era-aware image), and DoP (image-to-video, 5 seconds, 720–1080p, 100+ motion presets). Soul understands subcultural codes and micro-trends natively; DoP accepts short command-style prompts — one action plus one camera move at a time.
What's in the Higgsfield family
Soul 2.0 — image model up to 4K with T2I and I2I modes, 20+ style presets, and a Soul ID system for character consistency. Understands fashion terminology, subcultural language, micro-trends («phone flash», «disposable camera», «Y2K street»).
Soul Cinema — a separate image model focused on cinematic keyframes. Deep textures, film grain, era styling (70s, 90s). Suited for storyboards and concept art.
DoP — image-to-video model: 5 seconds, 720–1080p, 100+ motion presets (360 Orbit, Dolly In/Out, Whip Pan, FPV Drone, Bullet Time). Speed tiers: Lite (fast preview), Turbo (1.5–2× faster than Standard), Standard (max quality).
- Soul 2.0: T2I + I2I, up to 4K, 20+ style presets
- Soul Cinema: era-aware image (70s, 90s), film-grain styling
- DoP: I2V, 5 seconds, 720p–1080p, 100+ motion presets
- Soul ID — character consistency across multiple generations
- Face/Character Swap — visual tools, no text prompts
Soul prompt structure
A Soul 2.0 / Soul Cinema prompt is a creative brief, not a tag list. Formula: [Mood/Aesthetic] + [Subject with visual cues] + [References/Style] + [Technical hints].
Soul natively understands subcultural and aesthetic language: «phone flash editorial», «disposable camera Y2K street», «90s film grain», «editorial fashion forward». No need to over-explain these terms — the model recognizes the codes.
Length: short-to-medium prompts. Picking a style preset lowers the need for a complex prompt. For I2I (Soul Reference) use the reference image as a stylistic anchor — the prompt describes direction, not the visual.
DoP prompt structure (video)
DoP accepts short, direct prompts. Formula: [Active verb action] + [Camera direction]. One action plus one camera move at a time.
Use active verbs: «darts through», «leaps across», «crashes into», «glides», «spins». Drop fillers — command style beats descriptive style here.
Key rule: task separation. Lighting and framing go in the image prompt (Soul). Motion goes in the video prompt (DoP). Lighting in the video prompt causes flicker. Motion preset is selected SEPARATELY from the text prompt — it's a parameter, not part of the prompt.
«Shot on full-frame cinema camera» in the video prompt boosts realism.
Soul ID, HEX, Reference
Soul ID — character consistency across multiple generations. Use fresh, uniform reference photos. The main antipattern is mixing old and new references — this causes identity drift.
Soul HEX — automatic extraction and application of color palettes from reference images. Useful for brand series and moodboards with a fixed color scheme.
Soul Reference — I2I with visual anchoring: upload a reference, the prompt describes what to change or how to stylize. A pure white background on the input image is an antipattern — there's not enough depth for Soul.
Common mistakes
1. Long complex prompts for DoP
DoP is image-to-video with a focus on motion. Long descriptive prompts with story and emotion work poorly. Short command-style «Active verb + Camera direction» is the norm. If complexity is needed, push it into the image prompt (Soul), not into DoP.
2. Lighting and lens in the video prompt
Lighting and framing in a DoP prompt cause flicker. These parameters should be fixed at the image stage (Soul). The video prompt describes only motion and action. This is the basic task-separation rule between Soul and DoP.
3. Mixing old and new references for Soul ID
Soul ID delivers character consistency across multiple generations with fresh, uniform reference photos. Mixing old shots (different haircut, lighting, age) with new ones causes identity drift — the model averages the face. Use 3–5 fresh photos in one consistent style.
4. Pure white background on the input image
Soul (both T2I and I2I) works worse with a pure white input background — not enough depth or context for generation. Use a cluttered scene, a toned studio backdrop, or an environmental background. If the background must be neutral, describe it as «soft warm grey backdrop», not «pure white».
5. Prompting Face Swap / Character Swap
Face Swap and Character Swap (Recast) are visual tools, no text prompts. They run on input images and presets. Trying to edit a face or look via text in these modes won't work. Use Soul Reference (I2I) for text-driven editing or Face Swap for visual transfer.
Before / after examples
Example 1
Before
beautiful girl in 90s style
After
Phone flash editorial portrait of a young woman with bleached blonde hair and pink eyeshadow, oversized denim jacket, holding a disposable camera, Y2K street aesthetic, slight motion blur, harsh frontal flash, low fashion editorial mood.
Soul 2.0 natively understands «phone flash», «disposable camera», «Y2K» — no need to over-explain. Concrete aesthetic codes + subject with visual cues + technique.
Example 2
Before
the woman walks through the city looking sad and then turns around to look at the camera with cinematic lighting and atmospheric mood
After
Walks briskly through neon-lit alley, Whip Pan toward camera.
A DoP prompt should be SHORT. One action plus one camera move. Lighting and framing belong in the Soul prompt, not in DoP. Motion preset (Whip Pan) is a parameter, selected separately.
Example 3
Before
white background portrait, simple, clean
After
Editorial portrait of a young woman with long dark hair, wearing a cream cashmere sweater, against a soft warm grey studio backdrop, natural window light from the left, fashion-forward mood, shallow depth of field.
A pure white background on the input image is a Soul antipattern — not enough depth. A concrete background (warm grey backdrop) and lighting give the model something to anchor on.
