An identity-preserving reference-to-video LoRA for LTX-2.3 (22B). Give it a reference
photo of a person + a text prompt, and it generates a video that keeps that person's identity.
Built with overlap reference conditioning + TASS-RoPE (source-phase RoPE) and a differentiable
ArcFace identity loss. Runs in ComfyUI via the companion BFS Nodes.
Status: two reference modes are now available from the same recipe: the original
close-up face reference, and a newer character-sheet reference (face + full-body
views) that also carries clothing/appearance consistency — see
Character-Sheet Reference below. Other experimental
variants (native Gemma-vision conditioning, timestep-split texture injection) may be
released later if they prove out.
🎬 Examples
🎬 Face ID Base
🎬 Character Sheet
What it does
Reference-to-video (ref_t2v): one reference image → video of that identity performing the
prompt's action.
Identity is injected by placing the reference latent in the target's frame-0 RoPE grid
(overlap) and tagging it with a distinct source phase so the model knows it is a reference,
not the first frame to generate.
An auxiliary ArcFace face-similarity loss on the decoded prediction sharpens the identity.
How it works (technique)
Overlap reference + TASS-RoPE (source-phase)
The reference latent is concatenated to the video sequence sharing the frame-0 grid (classic
IC-LoRA "overlap"). To stop the reference from leaking into / being confused with the generated
first frame, each source gets a distinct multiplicative RoPE phase:
This "source tag" lets the model separate who is who in the sequence and strongly improves
identity transfer. Because the tag is positional, the same mechanism generalizes to multiple
references (source_id = 2, 3, 4, …) for multi-subject conditioning.
TASS-RoPE (Temporal-Adjacent Spatial-Shifted RoPE) is from ST-DRC, Chen et al.,
Spatial-Temporal Decoupled Reference Conditioning for Identity-Preserving Text-to-Video
Generation — arXiv:2606.02441. This LoRA uses an
overlap + per-source RoPE-phase variant of that idea.
ArcFace identity loss
During training the flow-matching prediction is decoded to pixels, the face is aligned and passed
through ArcFace (buffalo_l w600k_r50), and a cosine identity loss pulls the generated face toward
the reference embedding — the same recognition space used to evaluate identity.
Character-Sheet Reference (NEW)
The base model above (and the guidance further down) was trained on close-up face crops —
great for identity, but the model never saw the person's actual clothing or body, so it has to
invent an outfit rather than reproduce a specific one.
We've since trained a continuation checkpoint on a character-sheet reference instead: a
single image with 4 panels side by side — a face close-up, then full-body front / side / back
views of the same person, all on a white background. It uses the exact same mechanism
(source_id=2, layout=overlap) — it's just a different, richer reference image, so the model
learns to reproduce the reference's clothing and build consistently, not only the face.
This checkpoint continues training from the close-up FaceID LoRA above rather than replacing
it — in our tests the combination (close-up identity foundation + character-sheet continuation)
gave noticeably better and more consistent results than either recipe trained alone. If you only
have a plain headshot, the original close-up model still works well; the character-sheet
checkpoint is for when you want the generated video to also match a specific outfit/build.
File:Best_FaceID_CharacterSheet_v1.0_LoRA.safetensors
— use it exactly like the base LoRA above (same node, same source_id=2/layout=overlap), just
swap the reference image for a 4-panel character sheet instead of a headshot, and set
ref_resize_mode = native_resolution (see below). This checkpoint has no ArcFace projector —
its identity/appearance comes entirely from the overlap reference latent, so leave that dropdown
on None.
💡 Identity coming out weak? Load the base Best_FaceID_v1.0_LoRA.safetensors
alongside this one at a low strength (~0.2 or more) — mixing in some of the base close-up
LoRA strengthens identity while the character-sheet LoRA keeps handling clothing/body.
How to build a character sheet (for now): take (or generate) a normal reference photo, then
use ChatGPT (or another image-editing model, e.g. Nano-Banana) with this prompt:
"Convert the person in the photo into a character reference sheet with four panels arranged
left to right: a close-up of the face, then full-body front, side, and back views, on a white
background."
🛠️ We're training a dedicated Flux Klein 9B LoRA to generate these character sheets
directly (more consistent, no API needed) — still in training, not released yet. Until it's
out, use the prompt above with ChatGPT/gpt-image for the most faithful character sheet.
Usage difference vs. the close-up model: the character-sheet reference must be exactly
1536×1024 — that's the fixed resolution it was trained at, independent of the output video's
resolution. When using the LTX Identity Transfer node with this checkpoint, set
ref_resize_mode = native_resolution (not the default match_target) — otherwise the node
resizes the sheet down to the video's resolution and destroys the facial/clothing detail the model
relies on. This setting doesn't matter for the plain close-up reference, where ref and video share
the same resolution bucket.
same recipe, continued from the close-up checkpoint on 4-panel character-sheet references (face + front/side/back)
Reference image — what works best
This model was trained on face-focused reference images. Looking at the training data, every reference follows the same pattern:
Close-up / bust crop — roughly chest-up, with the face large and clearly visible.
A single subject, centered in the frame.
Frontal or near-frontal (slight 3/4 angles are fine, but the face is always well exposed).
Tightly cropped on the face/upper body — the training crops are about 460×406 (nearly square).
Recommendation: for the best identity transfer, give the model a reference that looks like this — crop your image to the face / upper body. Full-body shots, wide shots, or images where the face is small and far away work noticeably worse, because the model never saw references like that during training.
In short: a clean, frontal, well-lit close-up of the face is the ideal reference.
Using a full-body reference instead? See Character-Sheet Reference
above — the close-up guidance on this page is specifically for the base checkpoint.
Captions (IMPORTANT)
This model was trained on captions in the ref_t2v: format that describe the scene and the
person's action (not just "a person"). Identity in LTX is strongly prompt-driven — the more the
prompt describes, the better the identity holds. For best results:
Prefix prompts with ref_t2v:.
Describe the action, setting, framing, camera in present-progressive.
Describing identity attributes of the reference person (skin tone, hair, eyes, facial hair,
glasses, face shape) noticeably improves the result.
The shared ComfyUI workflow includes a Prompt Enhancer that looks at the reference image (via
the model's own multimodal Gemma-3 text encoder) and automatically enriches your prompt with the
person's identity attributes — so you don't have to describe them by hand.
Example (enhanced):
ref_t2v: A light-skinned man with long dark-brown hair past his shoulders, narrow rectangular
metal-frame glasses, light blue-gray eyes and light stubble is folding clothes in a laundry room,
medium-wide shot. He places the folded clothes into a basket, then carries it across the room.
Neutral indoor lighting, simple domestic environment.
Install ComfyUI-BFSNodes (ComfyUI Manager → Install via Git URL, or clone into custom_nodes/).
Dependencies (insightface, transformers, …) install automatically.
Load LTX-2.3 (checkpoint + the Gemma-3 text encoder / CLIP) as usual.
Add the LTX Identity Transfer (overlap + source-phase) node; feed it the reference image.
Using the character-sheet checkpoint? Set ref_resize_mode = native_resolution (see
Character-Sheet Reference).
Load this LoRA on the MODEL path.
Use the shared workflow (it wires the Prompt Enhancer → identity node → sampler).
A ready-made workflow JSON is included alongside the model.
About the ArcFace projector (optional — you don't need it)
The identity node has an optional ArcFace projector dropdown (it projects a face embedding into
extra conditioning tokens). You can leave it on None. In our experiments the projector's
effect was marginal — identity is carried almost entirely by the overlap reference latent (and
the prompt), so the projector adds very little on top and is not required for this model. It is kept
only as an optional extra; skipping it keeps the graph simpler with no meaningful quality loss.
Tips for best results
Use a clean, frontal, well-lit close-up as the reference — identity quality tracks reference
quality closely. Need clothing/body consistency too? Use the
character-sheet checkpoint instead of a plain close-up.
Let the Prompt Enhancer add the identity attributes (or write them yourself).
Keep the reference framing simple; busy/occluded reference faces degrade identity.
Limitations
First-frame / "mask" tendency: because identity comes from the raw reference latent, the
reference appearance (and framing / accessories like glasses) can partially copy into the video —
it can look like the reference is "pasted"/masked rather than freshly generated. Using a clean
frontal reference and the trim-first-frame node mitigates this.
Prompt dependence: identity is much stronger when the prompt describes the person. A bare,
generic prompt gives weaker identity — use the Prompt Enhancer.
Close-up bias (base checkpoint): the base training data skews to close-up / frontal faces, so
full-body or large-angle shots hold identity less well with the base checkpoint — use the
character-sheet checkpoint for those cases instead.
Metric caveat: ArcFace similarity is unreliable on small / turned / occluded faces — judge
visually, not only by score.
Citation
If you use this model, please cite the base model and the technique:
1@article{chen2026stdrc,
2 title = {Spatial-Temporal Decoupled Reference Conditioning for Identity-Preserving Text-to-Video Generation},
3 author = {Chen, Yuheng and Hu, Teng and Wang, Yuji and He, Qingdong and Ma, Lizhuang and Zhang, Jiangning},
4 journal = {arXiv preprint arXiv:2606.02441},
5 year = {2026}
6}
Identity loss: ArcFace (Deng et al., 2019) via InsightFace buffalo_l.
