How to Use Higgsfield Soul 2.0 for Character Videos
How to access Higgsfield Soul 2.0 and lock a consistent character across shots, from Higgsfield's own app to the 8frame canvas. Identity-lock workflow and cost.
If you need the same person to appear, recognizably, across a dozen clips, Higgsfield Soul 2.0 is the model built for it. This guide covers how to access it, both through Higgsfield's own app and directly on the 8frame canvas, and how to actually lock a character's identity so a face holds across every shot, camera angle, and lighting change.
What you'll end up with is a persistent character you can put in scene after scene without the face melting or drifting between generations. That consistency is the single hardest problem in AI video, and it's the reason a recurring-spokesperson series or a multi-scene narrative is even possible without a real actor and a real shoot.
What Higgsfield Soul is
Higgsfield Soul 2.0 is a character-consistency model. Its Soul ID system trains a personalized identity on a set of reference photos, learning a person's facial features, proportions, skin tone, and hair, then preserves that identity across every generation regardless of style, angle, or prompt. For the full definition, see What is Higgsfield. The underlying idea is what we call identity locking, the sibling of general character consistency: instead of hoping a prompt keeps a face stable, you pin the identity as a hard constraint.
Official access route
Higgsfield distributes Soul through its own platform.
The Higgsfield app. Go to higgsfield.ai, open the Character tool, choose the Soul ID Character option, and select Create. You upload 20-plus reference photos of the same person, ideally up to 70, shot in consistent lighting from different angles. More images improve pose range and consistency. Higgsfield trains the identity and returns a persistent token (something like soul-abc123) that encodes the character. Once trained, that identity carries across every image and clip you generate.
Subscription. Access is subscription-based. Higgsfield's plans bundle Soul ID with a large library of video and image models under one monthly fee, with tiers scaling by monthly credits. It's a strong package if you're all-in on Higgsfield's ecosystem and generate heavily every month.
The first-party app is the most feature-complete home for Soul ID, including its widest set of camera presets. The friction is the subscription-and-credits model: you're paying a recurring fee and drawing down a monthly credit pool whether or not this month's workload justifies it, and the character work lives on a different surface than the rest of your pipeline.
How to use Higgsfield Soul 2.0 on the 8frame canvas
On 8frame, Higgsfield Soul 2.0 is a model on the canvas, billed per clip, sitting next to the image models that make your reference stills and the upscalers that finish them. 8frame has no library of ready-made talking avatars; character work runs through Soul 2.0's identity lock from your own reference images, which is the honest and more flexible path.
-
Gather clean reference images. Pull together a consistent set of the person: even lighting, multiple angles, neutral expression range. If you don't have photos, generate a consistent character with Nano Banana Pro first and use those as references.
-
Add a Higgsfield Soul node and load the references. Wire your reference images into the node so the model locks the identity. This is the step that keeps the face stable across everything downstream.
-
Prompt the scene, not the face. With identity locked, your prompt describes the shot: setting, wardrobe, camera, mood. Example: "the character sits at a cafe window, speaks to camera, soft daylight, shallow depth of field, 9:16." For tested structures, see Higgsfield Soul 2 prompts for character stories and prompts for talking heads.
-
Generate a series and keep continuity. Because the identity is pinned, you can generate 10 different scenes and the same person shows up in all of them. Review on the canvas, upscale the winners with Topaz, and you have a coherent multi-shot set. A 5-second 1080p clip runs about 75 seconds.
Per-clip cost on 8frame is $0.30 to $0.50 for Higgsfield Soul 2.0, 1080p native, with no monthly subscription.
Official vs canvas: which to pick
| Your situation | Best route |
|---|---|
| You want the deepest Soul ID feature set and camera presets | Higgsfield app |
| You generate character content heavily every month | Higgsfield subscription |
| You want per-clip billing, no monthly commitment | 8frame canvas |
| You want the character work next to image + upscale nodes | 8frame canvas |
| You need to compare character work across video models | 8frame canvas |
| You're building reference stills from scratch first | 8frame canvas (Nano Banana into Soul) |
Cost math
Say you need a recurring spokesperson series: one character across 12 short scenes, each 5 seconds.
- On 8frame: 12 clips x roughly $0.40 = about $4.80 in compute, plus a few cents per reference still if you generate the character with Nano Banana Pro first. Upscale the keepers and you're near $6 total, with no subscription running in the background.
- Higgsfield subscription: the model is bundled in a monthly plan, which is efficient if you're producing character work constantly, and less efficient if this 12-clip series is a one-off that month.
- A real shoot with an actor for 12 setups is a day rate plus crew, comfortably into four figures, and you can't add scene 13 next week without booking them again.
The subscription wins on heavy, continuous use. The canvas wins on flexible, per-clip work and on keeping the character pipeline in the same place as everything that feeds it. For a broader look at how consistency workflows come together, our character consistency posts go deeper.
FAQ
How many photos do I need for Higgsfield Soul ID?
Upload at least 20 photos of the same person, and up to about 70 for best results. Use consistent lighting and capture the face from several angles. More images widen the pose range the model can reproduce and tighten consistency across generations.
Does 8frame have ready-made avatars?
No. 8frame has no library of pre-built talking avatars. Character work runs through Higgsfield Soul 2.0's identity lock, where you supply reference images of the person you want. That is more flexible than a fixed avatar library because the character is yours, not picked from a catalog.
How is Higgsfield different from a talking-avatar tool?
Avatar tools like the HeyGen class give you a pre-built presenter reading a script, which is fast when you want a stock spokesperson. Higgsfield Soul locks a specific identity you provide and places it in any scene, camera move, or story you prompt, which is the path when you need a custom character and cinematic shots rather than a template read.
Can I keep the same character across different video models?
The identity lock is Higgsfield Soul's own. On the 8frame canvas you generate the character with Soul 2.0 and can then route those frames into other models for effects or upscaling, but the consistency guarantee comes from Soul ID. Keep the character generation on Soul for the tightest face fidelity.
Need one character across a whole series? Lock an identity with Higgsfield Soul 2.0 on the 8frame canvas and generate every scene per clip. See the full model comparison for where character work fits your stack.