Sign In

Prototype by EasyWan22

1

Nov 8, 2026

(Updated: 7 days ago)

workflows
Prototype by EasyWan22

That was a 27-minute struggle for 2 seconds animation (On top of that, spent about 5 hours on the initial setup for easywan22).

Even with an RTX 5090, the 14B model of WAN 2.2 is quite a heavy-duty piece of software.

The specifications for this project are as follows:

1. Main Settings

Model:

WAN 2.2 I2V A14B

Model Format:

GGUF

Enable FastMix:

ON

Enable Base:

OFF

Boost1stStep:

ON

TorchCompile:

OFF

BlocksToSwap:

40

Size:

416px (input value)

Actual Resolution:

352 × 448

BaseFrameRate:

16 fps

Seconds:

2

NumFrames:

33

VFI:

OFF

Refiner:

OFF

2. Post-Processing and Output

Detailer:

OFF

Detailer Upscale:

OFF

UseUpscaler:

OFF

Enable Upscale:

OFF

PostProcess:

OFF

pointMosaic:

OFF

maskMosaic:

OFF

maskMosaicEndFrame:

999

addLabel:

ON

SnapStartEnd:

OFF

Save:

ON

3. LoRA and Sampling Settings

LoRA:

Wan22-I2V-A14B-4steps-lora-rank64-Seko-V1-H

LoRA Strength:

-1.6

Per-step CFG:

3.5 / 1.0 / 1.0

Sampling:

3 steps × 2 stages

Text Encoder:

Kijai (not confirmed)

4. Generation Results

Total Execution Time:

27:15

Generated Frames:

33

Resolution:

352 × 448

Video Duration:

Approx. 2 seconds

  • Stage 1 Sampling: 3 min 50 sec

  • Stage 2 Sampling: 1 min 12 sec

  • VAE Decoding: 14.85 sec

  • Final video successfully saved

Perhaps a safer approach would be to keep the current settings used to generate 33 frames, and then, during post-production video editing, duplicate each image so that it lasts for two frames.

Furthermore, by:

- Omitting intermediate frames of motion

- Using more pronounced, distinct poses

- Keeping the movement of hair and clothing subtle

- Keeping the camera static

...the result will look closer to traditional cel animation.

For the next attempt, aim to refine the settings to achieve a "hand-drawn animation style—animated on twos or threes" using WAN 2.2.

The version of EasyWan22 used includes post-processing features.

However, it is important to distinguish between the following:

- Video Frame Interpolation (VFI): Adding intermediate frames to create smoother motion.

- Frame Decimation (Thinning): Reducing the number of frames to achieve the "choppy" motion characteristic of traditional animation.

While EasyWan22’s standard post-processing suite includes VFI, have not found any specific settings for frame decimation—such as those required for "on-twos" (animating on every second frame) or "on-threes" (animating on every third frame).

Based on current setup, here is how would like to proceed next time:

1. Generate the video using WAN 2.2.

2. Enable post-processing.

3. Disable Video Frame Interpolation (VFI).

4. Apply frame decimation to achieve "on-twos" or "on-threes" timing.

In short, while it is possible to complete the entire process within the EasyWan22 workflow, the standard features alone are insufficient.

Moving forward, intend to integrate a frame-decimation function into the EasyWan22 workflow. want to find a solution that achieves this without increasing the load on the RTX 5090.

1

Comments