LTX-2.5 ComfyUI Workflows
The official Lightricks example graphs, explained — what each one does, which model files it needs, and how much VRAM it really takes.
Generation workflows
Text, image or audio in — video (and audio) out.
Text / Image to Video — Two Stage
Start here. Base-resolution pass, 2× spatial upscale, then a 3-step refine. Video and audio together.
- Conditioning:
- Text + optional first-frame image
- Best for:
- Default generation when you care about spatial detail
Text / Image to Video — Single Stage
Same inputs, one distilled pass, no upscaler. Faster and lighter; less spatial detail than two-stage.
- Conditioning:
- Text + optional first-frame image
- Best for:
- Fast previews and tighter VRAM
Audio to Video — Two Stage
Generate video that follows an input soundtrack. The original waveform is kept and muxed into the output.
- Conditioning:
- Audio (frozen) + optional image
- Best for:
- Video that must match a soundtrack, music or voice-over
Text to Audio — Single Stage
Audio only. No video VAE, no upscaler — the lightest graph in the set.
- Conditioning:
- Text
- Best for:
- Audio-only clips, SFX and ambience
Control & edit (IC-LoRA) workflows
Drive LTX-2.5 with a reference video, sheet, mask or motion tracks.
IC-LoRA Union Control (Depth / Canny / Pose)
Video-to-video driven by a depth, canny or pose annotator run on a reference clip. Depth is wired by default.
- Conditioning:
- Reference video (depth / canny / pose)
- Best for:
- Re-rendering a clip with new look while keeping motion and layout
Video to Video — IC-LoRA Edit
Use the source frames as IC-LoRA guides and change appearance with an identity / style / edit LoRA. Original audio is kept.
- Conditioning:
- Source video (+ frozen audio)
- Best for:
- Appearance edits on existing footage
IC-LoRA Ingredients (Reference Sheet)
Generate from one reference sheet of characters, props, wardrobe and locations. Describe each element by its place on the sheet.
- Conditioning:
- Reference sheet image
- Best for:
- Cast / props / location consistency from one image
IC-LoRA Motion Track
Image-to-video with sparse motion tracks you draw on the first frame. Direct exactly how objects move.
- Conditioning:
- Image + drawn tracks
- Best for:
- Directed motion from a still image
IC-LoRA Inpaint — Two Stage
Fill masked regions of a reference video. Source audio can stay frozen.
- Conditioning:
- Video + mask (+ frozen audio)
- Best for:
- Replacing or removing a region in existing footage
IC-LoRA Outpaint — Two Stage
Extend the canvas of a reference video — turn 9:16 into 16:9, or add headroom. Same LoRA as inpaint.
- Conditioning:
- Video + mask (+ frozen audio)
- Best for:
- Changing aspect ratio or extending the frame
VRAM quick check
Which transformer build to pick for your card, and whether it will fit at all.
| VRAM | Example cards | Verdict | Transformer | Advice |
|---|---|---|---|---|
| Under 16 GB | RTX 3060 12G, 4060, 4070, most laptops | Use online | nvfp4 | The 22B transformer alone is 18.7 GB even at NVFP4. Expect out-of-memory errors or extreme offloading. Generate online instead — same LTX-2 model, no setup. |
| 16 GB | RTX 4060 Ti 16G, 4080, 5070 Ti, A4000 | Tight | nvfp4 | Borderline. Use the NVFP4 transformer (Blackwell) or INT8 with aggressive offloading (--reserve-vram), the INT8 text encoder, single-stage graphs and low resolution. Slow, and not every graph will fit. |
| 24 GB | RTX 3090, 4090, A5000 | OK | int8 | Workable with the INT8 transformer + INT8 text encoder and the low-VRAM loader nodes. Two-stage graphs run; keep decode tiles small. Leave bf16 files for 32 GB+ cards. |
| 32 GB | RTX 5090, V100 32G | Comfortable | bf16 | Lightricks' stated minimum. bf16 transformer with the low-VRAM loaders fits; INT8 text encoder frees headroom for larger decode tiles and longer clips. |
| 48 GB+ | RTX 6000 Ada, A6000, H100 | Comfortable | bf16 | Everything in bf16 without offloading. Increase the decode tile size (see the Decode notes in each graph) for faster runs. |
Lightricks' official requirement is 32 GB+ VRAM. Lower tiers are community guidance for quantized builds, not a guarantee.
Under 24 GB?
The same LTX-2 model runs in our browser generator — 4K with audio, no setup, free credits to start.
Common errors & fixes
The five things that go wrong most often with LTX-2 in ComfyUI.
Legacy LTX-2.3 workflows
The 2.3 graphs still work with 2.3 model files; LTX-2.5 is the current release and the one we recommend.
T2V / I2V single stage (distilled or full) LTX-2.3_T2V_I2V_Single_Stage_Distilled_Full.json | Download JSON GitHub |
T2V / I2V two stage (with upsampling) LTX-2.3_T2V_I2V_Two_Stage_Distilled.json | Download JSON GitHub |
IC-LoRA union control (depth / pose / edges) LTX-2.3_ICLoRA_Union_Control_Distilled.json | Download JSON GitHub |
IC-LoRA motion tracking LTX-2.3_ICLoRA_Motion_Track_Distilled.json | Download JSON GitHub |
IC-LoRA HDR LTX-2.3_ICLoRA_HDR_Distilled.json | Download JSON GitHub |
IC-LoRA Dub-It (two stage) LTX-2.3_ICLoRA_DubIt_Two_Stage_Distilled.json | Download JSON GitHub |
IC-LoRA pixel spatial upscaler LTX-2.3_ICLoRA_Pixel_Spatial_Upscaler_Distilled.json | Download JSON GitHub |
IC-LoRA ingredients (reference sheet) LTX-2.3_ICLoRA_Ingredients_Single_Stage_Distilled.json | Download JSON GitHub |
IC-LoRA inpaint (two stage) LTX-2.3_ICLoRA_Inpaint_Two_Stage_Distilled.json | Download JSON GitHub |
IC-LoRA outpaint (two stage) LTX-2.3_ICLoRA_Outpaint_Two_Stage_Distilled.json | Download JSON GitHub |
V2V IC-LoRA single stage LTX-2.3_V2V_ICLoRA_Single_Stage_Distilled.json | Download JSON GitHub |
Text to audio single stage LTX-2.3_T2A_Single_Stage_Distilled.json | Download JSON GitHub |
Looking for model files?
Every transformer, text encoder, VAE, upscaler and IC-LoRA the workflows reference — with sizes, folders and Hugging Face links.
Browse the model list