LTX 2.3 Finetuned V2 FaceID

Advanced cinematic generation with native prompt enhancement and absolute identity control.

All generated content must follow Huggingface policy.

reference image

reference video (overrides image for v2v)

prompt (concept)

negative prompt

seed (-1 = random)

target megapixels

snap multiple

32 64

length (frames, /24 = seconds)

48 960

fps

8 60

input mode

i2v r2v v2a

conditioning fidelity

0.4 1

zerogpu override (0 = auto)

0 120

dmd strength

0 2

checkpoint

v1.2 v1.4

distillation method

dmd condsafe

skip upscaling (fast test)

evict text encoder (condsafe, solves OOM)

vbvr lora (video)

0 1

vbvr lora (audio)

0 1

dreamly lora (video)

0 1

dreamly lora (audio)

0 1

Native Reference Conditioning

enable for first pass

enable for upscale pass

Face-ID (IP-Adapter)

enable for first pass

enable for upscale pass

enable audio reference

reference audio (~4s clip)

identity guidance scale

0 10

isolate voice (stem separation)

normalize reference audio

output

·

Built with Gradio logo

·

·