The last few weeks, we’ve been tinkering with Wan 2.2 + Fun Control and figuring out how to drop it into our pipeline. After months on Wan 2.1, 2.2 feels noticeably better at “locking in” identity and shape. There is less drift, and more consistency.

Render stats: this 1:39 clip took ~3 hours on a 4070 (12 GB VRAM) to generate about 20 clips. There was some trial and error. Here’s what worked for us.

Frame rate vs. frames per clip

We switched to 15 fps with 90 generated frames per pass.

• That gave us ~6 seconds per batch, which let us build longer segments.

• Our rig is happiest around 81 frames; we pushed to 90 anyway to squeeze out more runtime.

• If you stick to 24 fps, the same frame budget only nets you ~3–3.5 seconds per batch, so your shots get choppier to stitch.

Lower OpenPose strength (to stop bleed-through)

Because the shot slowly pushes in, the Control Video started bleeding through and we got those white dot artifacts on Benny's face.

• Dialing down the OpenPose blend strength in the final Control Video helped.

• Close-ups: 0.25 worked well.

• Wider shots: going below 0.30 messed up the eyes. It’s a balance.

Note: you can still spot a few dots if you look closely in the demo—we left them in for funsies.

Test the first and last frames

We didn't notice the issue mentioned above until halfway through.

So to catch mid-clip artifacts early, we started proofing the first 5 and last 5 frames of every 90-frame set. If dots show up there, they’ll usually pop mid-clip too.

Use the same first frame for every batch

Even with the push-in, using the same exact starting frame for all batches kept the face consistent across clips.

• We first tried using the last frame of the previous batch to “match” into the next batch. That actually made the output drift/cartoonify over time, and the cut from last to first was obvious (Benny didn’t look the same).

• Reusing the same start frame every time fixed it—even if expression or camera position didn’t match perfectly. Wan 2.2 figured it out.

Stagger frame starts (for cleaner blends)

We staggered starts by 10 frames to overlap batches:

• Clip 1: 0-90

• Clip 2:  80-170

• Clip 3:  160- 250

Why: the overlap makes it easier to blend/crossfade batches and avoid hard color/contrast jumps. Net effect: a steadier, mostly seamless 1:30 shot.

Conclusion

Wan 2.2 is clearly better at teeth and eye movement, and with our ComfyUI workflow it’s already usable—but we’ve still got a few knobs to dial in.

Next up: a full scene test—more complex lighting, locations, and camera moves—to see how the model behaves under real production pressure.

If you want to experiment with a Wan Fun Workflow, don't forget that members also get access to the Wan 2.1 Fun Control Workflow we released earlier this week!