UGC agencies: the talking still is not a HeyGen twin
Fabric/Veed vs V5 twin are different contracts.
Fabric/Veed vs V5 twin are different contracts.
A UGC agency that treats every talking face as "the avatar" will quote the wrong row, miss the enrollment, and then wonder why week three does not look like week one. VEED Fabric 1.0 is a mouth pass on a plate plus a track. HeyGen Avatar V5 is a digital twin enrolled from a real person, then driven by a script or audio. One is per-clip lipsync. The other is an identity you keep. They share a talking mouth. That is the whole overlap.
Fabric 1.0 is a mouth pass on a plate you already have
VEED Fabric 1.0 is content type lipsync. Categories: audio-to-lipsync and image-to-lipsync. requires_image is true. Audio is true. Max output: 720p. Credits: 40. Qualities: 480p and 720p. Frame rates: 24fps, 30fps. Motion levels: Low, Medium, High. The fal path wants an image_url and an audio_url. Length follows the track. There is no duration list on the spec.
That is a talking still. Bring the portrait. Bring the voice file. The mouth follows. The catalog line about "professional fabric and texture video generation" sits next to those categories; it does not replace them. Fabric will not design the set or a reusable host. It re-derives motion from the still you sent this time.
The door is AI lipsync. Order: lock the face still, lock the track, run Fabric, caption. Skipping the first two is inventing the scene inside a mouth pass. 720p is the ceiling.
Avatar V5 is an enrolled twin, not a still
HeyGen Avatar V5 is also lipsync, and that is where agencies get lazy. Categories: premade-avatars. Features: Natural, Premade avatars, Script or audio. requires_image is false. Audio is true. Max output: 4K. Qualities: 720p, 1080p, 4K. Credits: 50. Aspects include 16:9 and 9:16. No duration list; billing is per second with a 50-credit floor.
A twin is not a lipsync trick applied to one jpeg. You capture a real person once, the system builds a reusable replica, and later scripts come out of that identity: face, voice, mannerisms. Consistency is the product. A still-to-talk row re-samples the face every job. Approximate sameness is not a twin.
V5 is the contract when a founder or a named creator will front the channel weekly and you already have a replica grant. It is the wrong contract for a one-off UGC test face you will throw away on Friday.
Agency tests live on the still contract
Volume UGC is usually five faces times three scripts. That is fifteen mouth passes, not fifteen enrollments. Fabric is how you audition a plate. You do not book a capture session to find out the hook is wrong. Do not sell a Fabric still as "your digital twin" to a client who thinks they can change the wardrobe in a dashboard. The still is the wardrobe.
Consent is not optional on either row. A real person's photo plus synthesized speech is putting words in their mouth. A twin is a replica: the grant has to name face, voice, uses, and take-down. A standard photo release for a shoot does not stretch.
Quote the row, not "a talking UGC"
| Fabric 1.0 | Avatar V5 | |
|---|---|---|
| Object | Talking still | Enrolled twin |
| Inputs | Image + audio | Script or audio after capture |
| Image required | Yes | No |
| Ceiling | 720p | 4K |
| Catalog credits | 40 | 50 |
| Consistency | Approximate, per clip | Same identity every render |
If the brief is "this jpeg says this line once," you are on Fabric. If the brief is "this person, every Tuesday, no camera," you are on V5.
FAQ
Can I skip capture and just run V5 on a headshot?
Not as this slug is specced. V5 is a digital-twin / premade-avatar row: script or audio, image not required. A locked portrait plus a track is Fabric 1.0. Pick the row that matches the inputs you actually have.
Why is Fabric only 720p if the client asked for 4K talking UGC?
Because 720p is the max on this slug. Qualities are 480p and 720p. 4K talking output is a different card — V5 lists 4K. Do not prompt Fabric into a resolution it does not ship.
Is a Fabric still "close enough" to a twin if I reuse the same photo?
No. Fabric re-derives the face from that still each generate. Delivery drifts. A twin is trained once and reused. Reusing a jpeg is not enrollment.
Should every UGC retainer include a V5 capture?
Only if the same sanctioned person will front enough videos to justify capture, and you have a replica grant. One-off test faces stay on the still-plus-track contract. Quote that honestly.