What’s new
- Taste is built in, not bolted on: Higgsfield says this version bakes taste straight into how the model composes, lights and styles a shot, rather than applying a style pass after the fact.
- Train a character once, then call it by ID: Soul ID trains a reusable character from photos; later generations just reference that character's ID, with an adjustable strength for how much it drives the result — no need to re-describe the look each time.
- Presets are API-level too: the 20+ curated presets map to a style reference you can pass in, with its own adjustable strength, so you get a look without writing a prompt from scratch.
- Can pad out a thin prompt for you: there's a built-in auto-enhance toggle that expands a short prompt automatically.
What it’s good for
- Editorial-style portraits and magazine-cover-style images.
- Series where the same face needs to recur — virtual personas, campaign sets — driven by a trained character reference instead of re-describing it each time.
- UGC and lifestyle-style social and ad creative.
Prompting tips
- Let a preset carry the look, keep the prompt simple: once a style is set, naming the subject and setting is usually enough — no need to stack style adjectives.
- Turn on auto-enhance when you're not sure: if your own prompt is thin, the enhance toggle fills in detail for you.
- Character strength is a dial, not just a switch: when using a trained character reference, strength runs 0 to 1 — push it up for a stronger likeness, or down to give the scene more freedom.
- Save the seed for a result you like: seed runs 1 to 1,000,000; the same seed with the same prompt gets you back to something close to a result you liked.
- A vague prompt still gives a different result each time: batch size tops out at 4, not dozens to sift through, so a clear prompt matters more than generating extra copies.
Editorial portrait, model in a beige tailored coat, standing on a city rooftop after rain, warm side backlight at sunset, 35mm film texture, cool tone, gazing off-camera.
Known limits
- No direct reference-image upload: there's no field for attaching a photo as a composition reference — keeping a character consistent means training a Soul ID character first and referencing it by ID.
- Batch size is only 1 or 4 — there's no 2 or 3 in between.
- Resolution is limited to two tiers, 720p and 1080p.
Using it in Nomi
On the generation canvas, add a image card, pick “Higgsfield Soul 2” in the model picker and choose a mode (Text to image). Add your references, write the prompt and generate. In the storyboard you can pick it per shot. The first time, connect any one of the providers above under Settings → Models.
