What is Higgsfield Soul 2.0?

What is Higgsfield Soul 2.0?

Soul 2.0 is one of Higgsfield’s three Soul image models, and the one with the most style controls. Higgsfield calls it “a new foundation image model built for creative, fashion-aware, culture-native generation” (Soul 2.0 page).

The family has three models, per the Soul help article. Soul runs on ready-made presets. Soul 2.0 adds your own styles and colours. Soul Cinema has a film look built in and no presets.

Three controls set Soul 2.0 apart:

• Presets: ready-made looks in categories such as TikTok Core, Instagram Aesthetics and Beauty.
• Moodboards: a style you build from your own pictures. The help article suggests “20 or more images in one cohesive aesthetic without faces”.
• Soul HEX: upload a picture and Soul copies its colour palette onto your image. Useful when every ad must sit in brand colours.

One catch shapes how you work. “With a reference image, the prompt field becomes unavailable.” The picture becomes the whole brief, and only a Soul ID and a Soul HEX palette can join it.

Soul 2.0 offers seven ratios, from 9:16 to 16:9. 4:5 is missing, though Meta’s feed specs list 4:5 at 1440 by 1800 (Meta ad specs). Generate at 3:4 and crop, or extend the image afterwards.

Five of 52 public sources praise Soul’s photorealistic stills and Cinema Studio’s directed camera moves (Reddit, Trustpilot).

What is Soul ID?

Soul ID is a saved copy of one person’s face that every Soul model can reuse. Higgsfield’s words: “Train it once on 20 or more photos of one person (up to 80), and the same face appears across every generation” (Soul ID article).

Think of it as a casting file. You hire the face once, then dress it, move it and light it in any scene.

Soul ID has four limits, per the help article:

• Expect “clearly the same person” rather than a pixel-identical face. Strong style shifts and odd angles bring “small drift”.
• One Soul ID holds one person. For two or more recurring characters in one scene, Higgsfield points you to Elements.
• The trained face “is not exportable as a standalone file”. Your images and videos download as normal files; the character stays on Higgsfield.
• It carries into video. Soul characters “appear in Elements automatically”, so Seedance clips can use the same face.

That last point is why Soul ID matters for ads. One spokesperson can appear in a still for the feed, then speak in a Seedance clip for Reels.

Five of 52 public sources credit Soul ID and Elements with keeping characters and products consistent across shots (YouTube, Reddit). Three report moderation blocking prompts and even characters Higgsfield generated itself (Reddit, Reddit).

How to create a Soul ID character

Training takes “a few minutes”, per the help article. The photo set decides the result: “a smaller set of clean, varied photos outperforms a large set of inconsistent ones”.

1. Open Soul and pick “Train new character”. Higgsfield’s photo rules for Soul ID sit in its own guide, How do I use Soul to generate images?, so keep it open beside you.
2. Upload 20 to 80 photos of the same person. Vary the angles and expressions, keep the face well lit and add at least one full-height shot.
3. Leave out sunglasses, hats, scarves and group photos. Use pictures from the last 4 to 5 months if the person’s look has changed.
4. Start training, then name the character.
5. Pick it from the Character tab whenever you generate in Soul, Soul 2.0 or Soul Cinema.

How to Train Consistent AI Characters with Higgsfield Soul IDCode And Create

Still inconsistent? Higgsfield’s advice is to retrain on a stronger set: swap repeated poses and shadowed shots, and add a full-height photo.

You can also invent a face. The help article says to generate portraits of the new character first, then train a Soul ID on them. An invented face has no real person behind it, which changes the ad rules below. The full method, from casting sheet to talking clip, is in how to make a realistic AI UGC character on Higgsfield.

What Soul costs

Soul runs on Higgsfield’s normal plans, and the plan cards above show annual billing. Starter costs $15 a month, Plus $49 a month and Ultra $129 a month.

The pricing page’s compare table caps Soul ID generations by plan, and annual billing raises every paid cap:

Soul 2.0 itself is cheap in credits. At 0.12 credits an image, Starter’s 200 credits cover about 1,666 images (200 divided by 0.12). The older Soul model costs 0.25. The pricing table gives Soul Cinema no rate of its own.

Two details move the real cost. A batch of 4 bills 4 images (credits article). And “Some plans include a one-time pool of free generations for Soul models”, listed on each plan card. The Plus card lists 5,000 free Soul 2.0 generations (pricing page, 9 October 2026).

The Free plan watermarks every image. Every plan, credit pack and refund rule: Higgsfield pricing.

Can you use a real person’s face in ads?

Yes, with that person’s documented consent, and with the platform’s AI label where its rules call for one. Higgsfield’s terms put the whole duty on you.

Section 5.3 makes you warrant that “you have obtained all consents, releases, and permissions required under applicable law from each individual whose face or voice is included”. Section 4.2(b) asks the same for any “name, likeness, voice, or other identifiable attributes”, yourself included (Terms of Use, 26 July 2026).

Three more clauses matter once the image runs as an ad:

• Training. Under 4.4, your inputs and outputs “may be used by Company to train” its models. Only an Enterprise Agreement switches that off. A model’s photos, uploaded for a Soul ID, fall under that clause.
• No passing off. Section 5.2(xi) bars you from presenting “that any Output is human-generated when it is not”.
• Labels and marks. Section 5.5 asks you to disclose AI “Where required by applicable law” and never to remove the provenance marks Higgsfield may embed under 6.4.

Three of 52 user sources object to the terms. An earlier July 2026 version licensed uploaded faces to Higgsfield perpetually; the 26 July 2026 terms quoted here no longer carry that clause (YouTube, Reddit).

A real face and an invented face fall under different laws. New York’s synthetic performer law covers a figure “not recognizable as any identifiable natural performer”. Read that way, the law reaches a Soul ID you invented and leaves out one trained on a real model’s face (AI UGC legal guide).

A real model’s face falls under likeness laws instead, such as California Civil Code 3344.

Platforms add their own rules. TikTok asks for its AI label on content “completely AI-generated” (TikTok AI label). Meta’s rules depend on the ad category (Meta AI disclosure). Clause by clause, plan by plan: Higgsfield in the rights matrix.

Swapping a face into existing footage raises sharper questions: Higgsfield face swap and Recast.

Every Higgsfield guide, sorted by job, is in the Higgsfield hub →