Skip to content

Latest commit

 

History

History
261 lines (224 loc) · 20.1 KB

File metadata and controls

261 lines (224 loc) · 20.1 KB

Higgsfield Model Guide

Video Models — Head to Head

Model Realism Character Motion Style range Duration Aspect ratios* Resolutions* Audio Best for
Kling 3.0 ★★★★★ ★★★★★ ★★★★★ ★★★★☆ 3–15s 16:9, 9:16, 1:1 Cinematic, long, audio, multi-shot. mode std/pro/4k
Kling 3.0 Turbo ★★★★☆ ★★★★☆ ★★★★☆ ★★★★☆ 3–15s 16:9, 9:16, 1:1 720p, 1080p Fast T2V + single start-frame animation, budget Kling 3.0
Kling 3.0 Omni ★★★★★ ★★★★★ ★★★★★ ★★★★☆ 3–15s Video clone, storyboard control
Kling 3.0 Omni Edit ★★★★★ ★★★★★ ★★★★☆ 3–10s in Edit footage at 3.0 quality
Kling O1 Video (legacy) ★★★★★ ★★★★★ ★★★★☆ ★★★☆☆ 5–10s Multi-ref (7), start/end frame
Kling O1 Video Edit (legacy) ★★★★☆ ★★★★★ ★★★★★ 3–10s in Relight, restyle, swap, remove
Kling 3.0 Motion Control ★★★★★ ★★★★☆ ★★★★★ ★★★☆☆ 3–30s Optional Motion transfer from reference video
Kling 2.6 (legacy) ★★★★★ ★★★★★ ★★★★☆ ★★★☆☆ 5/10s 16:9, 9:16, 1:1 Character drama, realism; native audio via sound toggle (default on)
Kling 2.5 Turbo ★★★★☆ ★★★★☆ ★★★★☆ ★★★☆☆ 5–10s Fast Kling iteration
Kling 2.1 Master (deprecated) ★★★★☆ ★★★★☆ ★★★☆☆ ★★★☆☆ 5–10s Deprecated — removed from platform. Use Kling 2.6 or 3.0
Sora 2 (UI-only) ★★★★☆ ★★★☆☆ ★★★★★ ★★★★☆ 4–12s 720p (base), 1080p (Pro/Max tiers) Epic scale, physics, action, multi-shot + sound — UI-only, confirmed present in the UI 2026-07-06 as a 4-variant family: Sora 2 (720p) / Sora 2 Pro (1080p) / Sora 2 Max / Sora 2 Pro Max (both 1080p, "BY HIGGSFIELD" enhanced tiers), all 4–12s. Not in the API/MCP catalog — UI generations only
Wan 2.7 ★★★★★ ★★★★☆ ★★★★★ ★★★★★ 2–15s 16:9, 9:16, 1:1, 4:3, 3:4 720p, 1080p 60fps, T2V/I2V/R2V/edit, first+last frame
Wan 2.6 ★★★★☆ ★★★☆☆ ★★★★☆ ★★★★★ 5/10/15s 16:9, 9:16, 1:1 Artistic, stylized, improved physics
Wan 2.5 ★★★★☆ ★★★☆☆ ★★★★☆ ★★★★★ 5–10s Native audio, artistic, fantasy
Wan 2.5 Fast ★★★☆☆ ★★★☆☆ ★★★★☆ ★★★★★ 5–10s Fast Wan iteration with audio
Seedance 2.5 4–30s auto, 21:9, 16:9, 4:3, 1:1, 3:4, 9:16 480p, 720p Omni-reference: 30 images / 10 videos / 10 audio refs, 30s in one generation, plus video_edit and forward/backward video_extension modes. No 1080p/4K, no start/end-frame role, no genre hint. Not yet field-rated
Seedance 2.0 ★★★★★ ★★★★★ ★★★★★ ★★★★☆ 4–15s auto, 21:9, 16:9, 4:3, 1:1, 3:4, 9:16 480p, 720p, 1080p, 4k 12-asset multimodal, complex motion, native 4K (mode=std), genre hints
Seedance 2.0 Fast ★★★★☆ ★★★★☆ ★★★★★ ★★★★☆ 4–15s auto, 21:9, 16:9, 4:3, 1:1, 3:4, 9:16 480p, 720p mode=fast of Seedance 2.0 — cheaper/faster, no 1080p/4K, lower plan tier
Seedance 2.0 Mini ★★★★☆ ★★★★☆ ★★★★☆ ★★★★☆ 4–15s auto, 16:9, 9:16, 4:3, 3:4, 1:1, 21:9 480p, 720p Distinct API id (seedance_2_0_mini, added ~2026-06): budget tier with the full reference-input surface (image/video/audio refs) + native audio; no 1080p/4K
Seedance 1.5 Pro ★★★★☆ ★★★★☆ ★★★★☆ ★★★★☆ 4/8/12s auto, 16:9, 9:16, 4:3, 3:4, 1:1, 21:9 480p, 720p, 1080p Multilingual audio, lip-sync, drama
Seedance Pro (legacy UI label) ★★★☆☆ ★★★☆☆ ★★★☆☆ ★★★☆☆ 10s Legacy label — not in the API catalog (2026-07-05 snapshot); superseded by Seedance 1.5 Pro (seedance1_5) and Seedance 2.0 Fast/Mini
Veo 3.1 ★★★★★ ★★★★☆ ★★★★☆ ★★★★☆ 4/6/8s 16:9, 9:16 Ref images, first/last frame, extension, 4K
Veo 3.1 Fast ★★★★☆ ★★★★☆ ★★★★☆ ★★★★☆ 4/6/8s Fast iteration, same caps as 3.1
Veo 3 ★★★★☆ ★★★☆☆ ★★★★☆ ★★★☆☆ 4–8s 16:9, 9:16 Nature, environment, stable model
Veo 3.1 Lite ★★★★☆ ★★★★☆ ★★★★☆ ★★★★☆ 4/6/8s 16:9, 9:16, auto Budget 3.1 quality, 1080p, I2V, volume
Veo 3 Fast ★★★☆☆ ★★★☆☆ ★★★☆☆ ★★★☆☆ 8s Fast, stable, volume content
Gemini Omni Flash ★★★★☆ ★★★★☆ ★★★☆☆ ★★★☆☆ 4–10s 16:9, 9:16 720p Reference-driven video (image + video refs), native audio, fast social clips
Grok Imagine Video ★★★★☆ ★★★☆☆ ★★★★☆ ★★★★☆ 1–15s 16:9, 9:16, 1:1 Video editing, animate images, social clips
Grok Imagine 1.5 ★★★★☆ ★★★★☆ ★★★★☆ ★★★★☆ 2–15s 480p, 720p I2V-only preview — animate one start image, native audio direction
Minimax Hailuo 2.3 ★★★★★ ★★★★☆ ★★★★★ ★★★★☆ 6/10s 512, 768, 1080 VFX, fluid motion, anime, physics
Minimax Hailuo 2.3 Fast ★★★★☆ ★★★★☆ ★★★★★ ★★★★☆ 6/10s 512, 768, 1080 Fast iteration, batch creation
Minimax Hailuo 02 ★★★★☆ ★★★☆☆ ★★★★★ ★★★☆☆ 6/10s 512, 768, 1080 Dance, sports, fluid motion
Minimax Hailuo 02 Fast ★★★☆☆ ★★★☆☆ ★★★★☆ ★★★☆☆ 6/10s 512, 768, 1080 Budget motion, 512p
FLUX 3 Video 5–20s auto, 21:9, 2:1, 16:9, 4:3, 1:1, 3:4, 9:16 720p, 1080p T2V + multi-frame I2V + video continuation with synchronized audio; start/end frame + image/video reference roles; only model in the catalog with a native 2:1 ratio. Not yet field-rated
Higgsfield DoP (Lite/Standard/Turbo) ★★★☆☆ ★★★☆☆ ★★★★☆ ★★★☆☆ 3–5s I2V specialist, 50+ camera presets, optical physics

* Aspect ratios / Resolutions columns are sourced from the specs layer (specs/MODEL-SPECS.md, models_explore snapshot 2026-08-07) — do not hand-edit them. A means the model is legacy/unsnapshotted or the snapshot does not expose that field; verify live before promising values for those rows. Duration cells for snapshot-covered models are cross-checked against the specs by scripts/validate.py.

The catalog also lists utility/system entries — AutoSprite (game sprite-sheet animation), MS Image (Marketing Studio ad images), LLM text, upscalers, background removers, outpaint, Sync Lipsync 3 (audio-driven lipsync retiming). These are pipeline tools, not prompt-crafted generation models, and are intentionally out of scope for these tables. (Explainer Video left the catalog in the 2026-08-07 snapshot — it had already dropped out of the CLI list at 2026-08-01. Verify in the UI before referencing it.)

New in the 2026-08-07 snapshot — not yet field-rated (no star rows until real generations back them): Seedance 2.5 (seedance_2_5 — see the dedicated row above and skills/higgsfield-seedance-2-5/SKILL.md) and FLUX 3 Video (flux_3_video, Black Forest Labs — T2V / multi-frame I2V / continuation, 5–20s, 720p–1080p). Also changed: Grok Video 1.5 gained image + audio reference roles (it is no longer I2V-only), and the live CLI reports three mutual-exclusion rules on it — references cannot be combined with start_image, audio references require at least one image reference, and 1080p is unavailable once references are attached. MiniMax H3 gained a batch_size parameter (1–4). Enums per specs/MODEL-SPECS.md; verify credits in the UI before recommending.

New in the 2026-08-01 snapshot — not yet field-rated: MiniMax H3 (minimax_h3 — multimodal keyframes + image/video/audio references, 5–15s, 2K, incl. 21:9) and Happy Horse Video (happy_horse_video — T2V / single start-frame, 3–15s, 720p/1080p).

Image Models — Head to Head

Model Quality Faces Style range Speed Best for
Soul 2.0 ★★★★★ ★★★★★ ★★★☆☆ ★★★★☆ Fashion, portrait, aesthetic
Soul Cinema Preview ★★★★★ ★★★★★ ★★★☆☆ ★★★★☆ Cinematic keyframes, close-ups, film grain
Soul Cast ★★★★☆ ★★★★★ ★★★☆☆ ★★★☆☆ Consistent cinematic character identity (16:9, budget 10–500)
Soul Location ★★★★☆ ★★★☆☆ ★★★★☆ Environment / location generation, 9 aspect ratios incl. 21:9 + 9:21
Kling Image 3.0 ★★★★★ ★★★★★ ★★★★☆ ★★★★☆ Native 4K, series mode, storyboarding
Kling Image 3.0 Omni ★★★★★ ★★★★★ ★★★★☆ ★★★★☆ Advanced editing, strongest prompt fidelity
Nano Banana Pro ★★★★★ ★★★★☆ ★★★★☆ ★★★☆☆ Max fidelity, Thinking mode, 14 refs
Nano Banana 2 ★★★★★ ★★★★☆ ★★★★☆ ★★★★☆ Fast pro-quality, text rendering, consistency
Nano Banana 2 Lite ★★★★☆ ★★★★☆ ★★★★☆ ★★★★★ Budget NB2 — 1k only, thinking MINIMAL/HIGH
Seedream 4.5 ★★★★☆ ★★★★★ ★★★★★ ★★★★☆ Reference consistency, dense text, 4K
Seedream 5.0 Lite ★★★★☆ ★★★★☆ ★★★★★ ★★★★★ Reasoning, search, multi-output, layouts
GPT Image 2 ★★★★★ ★★★★☆ ★★★★☆ ★★★★☆ Native 4K, best text/typography, reasoning compositions
GPT Image 1.5 ★★★★☆ ★★★★☆ ★★★★☆ ★★★★☆ Text-in-image, instruction following
OpenAI Hazel ★★★★☆ ★★★★☆ ★★★★☆ ★★★★☆ Reference-based editing, best text rendering (quality low/medium/high)
Recraft 4.1 ★★★★☆ ★★★☆☆ ★★★★★ ★★★★☆ Logos/icons/vector, product mockups, hex brand palettes
Flux 2 ★★★★☆ ★★★★☆ ★★★★☆ ★★★★☆ High-quality FLUX generation (pro/flex/max)
Flux Kontext ★★★★☆ ★★★★☆ ★★★☆☆ ★★★★☆ Editing existing images

Decision Flowchart

Is this image or video?
├── IMAGE
│   ├── Is it a person / portrait? → Soul 2.0
│   ├── Need cinematic keyframe for I2V pipeline? → Soul Cinema Preview
│   ├── Same character identity across many shots? → Soul Cast
│   ├── Environment / location plate? → Soul Location
│   ├── Need native 4K / image series / storyboarding? → Kling Image 3.0
│   ├── Need maximum sharpness / 4K? → Nano Banana Pro
│   ├── Fast pro-quality / text rendering / consistency? → Nano Banana 2
│   ├── Need reference consistency or dense text? → Seedream 4.5
│   ├── Complex layout / real-time data / multi-panel output? → Seedream 5.0 Lite
│   └── Need to edit an existing image? → Flux Kontext
│
└── VIDEO
    ├── Do you want to EDIT existing footage?
    │   ├── Relight, restyle, swap, remove, transform? → Kling O1 Video Edit (Edit Video tab)
    │   └── Higher quality edit with 3.0 architecture? → Kling 3.0 Omni Edit
    │
    ├── Is a human character the focus?
    │   ├── Need audio, long clip (up to 15s), or multi-shot? → Kling 3.0
    │   ├── Need to clone from a reference video? → Kling 3.0 Omni
    │   ├── Need per-shot storyboard control? → Kling 3.0 Omni
    │   ├── Need multi-reference inputs (up to 7) or start/end frame? → Kling O1 Video
    │   ├── Fast iteration on Kling quality? → Kling 2.5 Turbo
    │   └── Standard length, audio optional (`sound` toggle)? → Kling 2.6
    │
    ├── Need motion transfer from reference video?
    │   └── → Kling 3.0 Motion Control
    │
    ├── Animate a still image with cinematic camera?
    │   └── → Higgsfield DoP (Lite/Standard/Turbo)
    │
    ├── Need very long camera motion (up to 30s)?
    │   └── → Kling 3.0 Motion Control
    │
    ├── Is the source an EXISTING video the user wants changed or continued?
    │   ├── Change one thing inside it (object, background, BGM, spoken
    │   │   language)? → Seedance 2.5 `video_edit`
    │   ├── Add footage after — or before — it? → Seedance 2.5
    │   │   `video_extension` (forward / backward)
    │   └── Restyle / VFX-transform the whole plate? → Seedance 2.0
    │       (skills/higgsfield-seedance-vfx) or Grok Imagine Video
    │
    ├── One clip longer than 15s, or more than 9 image refs?
    │   └── → Seedance 2.5 (4–30s, 30 image / 10 video / 10 audio refs) —
    │       accepting the 720p ceiling; finish on Seedance 2.0 if the
    │       deliverable needs 1080p/4K
    │
    ├── Is scale / spectacle the focus?
    │   └── Explosions, crowds, physics, epic landscapes? → Seedance 2.0 or
    │       Minimax Hailuo 2.3 (Sora 2 is UI-only — not in the API catalog
    │       as of 2026-07-05; verify in the live UI before using it)
    │
    ├── Is artistic style the priority?
    │   ├── 60fps, first+last frame, reference images? → Wan 2.7
    │   └── Fantasy, stylized, painterly, surreal? → Wan 2.5 / 2.6
    │
    ├── Is fluid motion / physical performance the focus?
    │   └── VFX, anime, fluid motion, physics? → Minimax Hailuo 2.3
    │   └── Dance, sports, martial arts? → Minimax Hailuo 02
    │   └── Video editing / restyle existing footage? → Grok Imagine Video
    │   └── Animate a still image (10s, audio)? → Grok Imagine Video
    │
    ├── Is the environment / nature the hero?
    │   ├── Subject consistency / reference images? → Veo 3.1
    │   ├── Define start + end frame? → Veo 3.1
    │   ├── Extend existing Veo video? → Veo 3.1
    │   ├── Fast iteration (Veo quality)? → Veo 3.1 Fast
    │   ├── Budget Veo 3.1 quality / volume? → Veo 3.1 Lite
    │   └── Weather, wildlife, landscape (stable)? → Veo 3
    │
    └── Need fast iteration / social content?
        ├── Audio required? → Seedance 1.5 Pro
        ├── 12-asset multimodal reference? → Seedance 2.0
        ├── Image + video refs with native audio (720p)? → Gemini Omni Flash
        └── No audio, speed first? → Seedance 2.0 Fast / Mini with
            generate_audio off ("Seedance Pro" is a legacy UI label —
            not in the API catalog)

Kling Generation vs Edit Mode

The Kling lineup has two distinct interfaces in Higgsfield:

Create Video tab — Generate new video from text, image, or references

  • Kling 3.0, 3.0 Omni, O1 Video, Motion Control, 2.6, 2.5 Turbo

Edit Video tab — Transform existing footage (video-to-video)

  • Kling O1 Video Edit: natural language edits, no masking, 720p output
  • Kling 3.0 Omni Edit: same concept at 3.0 quality tier

Key distinction: Edit mode preserves the original motion structure, camera path, and spatial relationships of your source footage. You're transforming the look/content, not the movement.

Edit Prompt Formula

Change [Target] to [New State], keep [everything else] unchanged

Examples:

  • "Change the lighting to golden hour dusk, keep character and motion unchanged"
  • "Restyle as 1970s film grain, keep all composition"
  • "Remove the background person, keep everything else exactly as is"

Model + Camera Control Compatibility

Some camera controls perform better on certain models:

Camera Control Best model
Dolly In (emotional close-up) Kling 2.6 / 3.0
FPV Drone (kinetic chase) Kling 2.6, Sora 2†
360 Orbit (character isolation) Kling 2.6 / 3.0
Crane Up (epic reveal) Sora 2†
Timelapse Landscape Veo 3
Hyperlapse Veo 3, Sora 2†
Handheld (documentary feel) Kling 2.6, Veo 3
Action Run (physical chase) Minimax Hailuo 2.3, Kling 2.6
Super Dolly Out (scale reveal) Sora 2†
Dutch Angle (horror/tension) Wan 2.5, Kling 2.6
Long camera motion path Kling 3.0 Motion Control
Motion transfer from reference Kling 3.0 Motion Control

† Sora 2 is UI-only — not in the API catalog as of 2026-07-05; verify in the live UI before recommending. Catalog-verified fallbacks for scale/physics shots: Seedance 2.0, Minimax Hailuo 2.3.


Model + Motion Preset Compatibility

Preset type Best model
Transformation (Werewolf, Cyborg, Animalization) Wan 2.5, Kling 2.6
Elemental (Fire, Water, Earth, Air) Wan 2.5, Sora 2†
Explosion / Destruction Sora 2†, Seedance 2.0
Surreal / Glitch / Multiverse Wan 2.5
Horror presets Kling 2.6, Wan 2.5
Dance / Motion glow Minimax Hailuo 2.3
Nature effects (Sakura, Bloom, Northern Lights) Veo 3, Wan 2.5
Bullet Time / Slow motion Kling 2.6, Sora 2†
Stylized (Anime, Pixar, Claymation) Kling 3.0, Wan 2.5

† Sora 2 is UI-only — not in the API catalog as of 2026-07-05; verify in the live UI before recommending (see the Camera Control note above).


Credit Cost Reference (approximate)

Model Credit cost per generation
Seedance Pro (legacy UI label — see video table) Low
Kling 2.5 Turbo Low–Medium
Wan 2.5 Low–Medium
Minimax Hailuo 2.3 Medium
Minimax Hailuo 02 Medium
Kling 2.6 Medium
Sora 2 (UI-only — verify live) Medium–High
Kling O1 Video Edit ~9 credits
Kling 3.0 ~10 credits
Kling 3.0 Motion Control Medium–High
Veo 3.1 Lite Medium
Veo 3 High
Wan 2.7 Low–Medium
Wan 2.5 Low–Medium

Plans: Free (25 credits/mo) · Basic $6/mo (150) · Pro $27/mo (700) · Ultimate $55/mo (1500)


Higgsfield DoP — Image-to-Video Specialist

Higgsfield DoP is Higgsfield's native Image-to-Video (I2V) system with three quality tiers:

  • Higgsfield Lite — 720p, 3–5s, fastest
  • Higgsfield Standard — 720p, 3–5s, balanced
  • Higgsfield Turbo — 720p, 3–5s, highest quality

What makes DoP unique:

  • Specialized I2V generative model — designed for animating static images
  • 50+ cinematic camera presets (Bullet Time, Crash Zoom, 360 Orbit, Robo Arm, FPV Drone, etc.)
  • Simulates real-world optical physics: bokeh, parallax, natural lighting shifts as camera moves through 3D-mapped environment
  • Multi-Frame Guidance: starting image + "last frame" reference for seamless transitions
  • Preset categories: All, New, Trending, Effects, Basic Camera Control, Epic Camera Control, Catch the Pulse, Mix

Best for: Social media ads, product showcases, film pre-visualization, animating hero images with cinematic camera movement