If you have ever tried to make an AI video model animate knitting, threading or netting, you probably know the feeling. The frames look beautiful. The fabric looks real. And the one thing the shot is actually about — thread moving, looping and building into a pattern — just doesn’t happen.
I ran into this wall again and again. I tried MiniMax. I tried Wan. I rewrote prompts, changed reference images and tried again. Across multiple attempts, neither of them could produce a usable knitting or threading animation.
Then, in November 2026, I tried LTX-2.5 inside ComfyUI. It did it.
Why this is a bigger problem than it sounds
Knitting and threading are not niche. A huge amount of commercial product work depends on exactly this kind of motion:
- Furniture shots — upholstery, woven chair seats, cane and rattan, rope details.
- Clothing and fashion — knitwear, stitching, fabric forming around a garment.
- Textiles and home products — rugs, nets, mesh, anything where the material is the story.
In these shots the motion is the product message: this is how it’s made, this is the craft. If the model can only give you a pretty still that wobbles, the shot is useless. And because most AI video comparisons test faces, cars and camera moves, nobody warns you about this before you’ve wasted days on it.
What I found
MiniMax and Wan: great frames, no knitting
Both are strong models for general cinematic shots. But when I asked them for threads that actually loop, cross and build into a knit or a net, they did not deliver. Multiple tries, same result. This is not a one-off bad generation — it was consistent enough that I stopped treating it as a prompting problem.
LTX-2.5 in ComfyUI: it actually threads
LTX-2.5 is Lightricks’ open-weights video model, released in August 2026 with native ComfyUI support. On the same kind of shot, it produced the threading and netting motion that the others could not. For this specific type of animation, it is the better option.
The practical takeaway
If your shot is about knitting, threading, weaving, netting, stitching or fabric forming on an object — furniture, clothing, anything textile — start with LTX-2.5 in ComfyUI, not MiniMax or Wan. You will save yourself the hours I lost.
For everything else, don’t pick a favourite model and stick to it. Each one has a type of motion it is good at and a type it simply cannot do. Testing the specific motion your shot needs, early, is the real skill.
This is based on my own repeated testing on real shots, not a lab benchmark. Models update fast, so if a newer MiniMax or Wan release handles this, I’ll update this page.
More on how I use AI video in a production pipeline →
I made a title sequence entirely with AI — what broke →
Questions people ask
Which AI video model can animate knitting or threading?
In my testing, LTX-2.5 running in ComfyUI was the only model that produced a usable knitting, threading or netting animation. MiniMax and Wan failed on the same kind of shot across multiple tries.
Can MiniMax or Wan do knitting animation?
Not in my tests. Both produce good-looking frames, but when the shot needs threads to actually loop, cross and build up into a knit or net, they did not deliver a usable result.
What is LTX-2.5?
LTX-2.5 is Lightricks' open-weights video model, released in August 2026 with native ComfyUI support. You can run it locally inside a ComfyUI workflow.
When should I use LTX instead of MiniMax or Wan?
Use LTX when the motion itself is the subject: knitting, weaving, threading, netting, stitching or fabric being built up, for example in furniture, upholstery, textile and clothing shots. For general cinematic shots, test all of them, because the best model depends on the shot.
Need AI-assisted product motion?
I combine AI video with 3D and traditional motion design for furniture, fashion and product shots. Open to full-time roles and select freelance projects — Dubai, the Gulf and remote.
Email me CV & availability See my work