Wan 2.7 Uncensored ref2v vs Wan 3.0 Spicy ref2v: Lock a Face

Key Takeaways

• Wan 3.0 Spicy ref2v takes far more reference material (up to 10 images/5 videos/5 audios vs. 3/3), but it dropped negative_prompt entirely; Wan 2.7 Uncensored ref2v still has it.
• Wan 2.7 Uncensored ref2v tops out at 15s on 720p/1080p only; Wan 3.0 Spicy ref2v runs to 30s across 480p/720p/1080p.
• Verified on the console today, both land on $0.09/s at 720p — a 10% promo off $0.10/s for Wan 2.7 Uncensored, a 33% promo off ~$0.134/s for Wan 3.0 Spicy. Coincidence, not policy — recheck before budgeting, since both are time-limited.

Wan 3.0 Spicy ref2v: A Character Consistency Guide covers that one endpoint alone. Seedance 2.5 Spicy vs Wan 3.0 Spicy: Multi-Character Fix solves a different problem — several characters in one frame. Wan 3.0 Spicy vs Wan 2.7 i2v Spicy: Should You Migrate? compares the i2v endpoints on migration grounds. This is a fourth, narrower question: for one character across several ref2v shots, which field helps more — negative_prompt, or a bigger reference set?

What Goes Into Each ref2v Call

Both take a reference set plus a prompt and return a new shot carrying that likeness. alibaba/wan-2.7-ref2v-uncensored: up to 3 images/3 videos, duration 2-15s, size 720p/1080p, plus seed and negative_prompt. alibaba/wan-3.0-ref2v-spicy: up to 10 images/5 videos/5 audios, duration 2-30s, size 480p/720p/1080p, six aspect_ratio presets, audio_enable, prompt_expansion_enable (default true) — no negative_prompt anywhere in its schema.

Spec by Spec

The Trade: More References, No negative_prompt

More reference material helps hold a face across shots that shift lighting, angle, or wardrobe — 8-10 stills give Wan 3.0 Spicy ref2v more to anchor to than 3 images allow. But it dropped negative_prompt, so there is no request-level way to say "not this hairstyle" when the reference set still drifts. Wan 2.7 Uncensored ref2v keeps that field, trading a smaller reference cap for a direct steering knob on what should not appear.

Price Per Second, Verified Today

Checked live on both console Pricing blocks, 2026-08-31: Wan 2.7 Uncensored ref2v's 10% promo puts 720p at $0.09/s (from $0.10/s), 1080p at $0.135/s. Wan 3.0 Spicy ref2v's deeper 33% promo puts 720p at $0.09/s too (from ~$0.134/s; 480p $0.045/s, 1080p $0.18/s). Same $0.09/s at 720p, right now — two unrelated, time-limited promos intersecting, not a shared tier, so recheck before budgeting. Billing on both: input reference length (capped 5s) plus output duration, at that rate.

Which One to Pick

Pick Wan 2.7 Uncensored ref2v when the look locks with 1-3 references and the risk is a specific drift negative_prompt can name. Pick Wan 3.0 Spicy ref2v when the character must survive bigger swings in pose, lighting, or wardrobe, where a larger reference set does more work than a negative prompt. A two-stage pipeline — establish on Wan 3.0 Spicy ref2v, then clean up drift on Wan 2.7 Uncensored ref2v with negative_prompt — works too. Both share one Siray API key with the catalog, so switching is a model-name change, not a new integration.

FAQ

Does Wan 3.0 Spicy ref2v have any way to exclude unwanted elements the way negative_prompt does?

No. Its schema has no negative_prompt field; the closest lever is a larger, cleaner reference set and a more specific prompt.

Which gives more control over a locked character: more reference images, or negative_prompt?

Different things. More references (Wan 3.0 Spicy) help the model generalize a likeness; negative_prompt (Wan 2.7 Uncensored) rules out a specific recurring drift. Neither replaces the other.

Can both endpoints be used in the same pipeline?

Yes — same reference input type, same billing formula, so one can establish the character and the other can clean up drift-prone shots.

Create your free Siray account and test both ref2v endpoints against the same reference set before committing to one.