🔁 This workflow is based on my existing I2V workflows and behaves almost identically.
▶️ You simply load a single image as both the start and end frame, enter your prompt, and run the workflow. That’s it.
📝 For general usage and guidance, I’ve added inline markdown notes directly inside the workflow.
There is only a KSampler Advanced version, as it performs virtually on par with the MoEKSampler. In practice, you won’t notice any meaningful difference in output quality.
There isn’t much more to explain, it’s intentionally straightforward.
💬 If you have questions or feedback, please use the comments rather than DMs.
That way, answers are visible to everyone and can help other users as well.
✨ Have fun. I’m genuinely excited to see what you create with it.
⚠️ Disclaimer: You are solely responsible for anything you generate using this workflow.
Description
FAQ
Comments (12)
error - cannot access local variable 'clip_vision_out" where it is not associated with a value
Thanks for reporting this issue. This workflow works without CLIP Vision - those inputs are optional and not used. Did you connect something to those inputs by accident?
What's your setup?
- ComfyUI version?
- Any error messages during workflow load?
It's likely an environment-specific issue. Please provide some more information. Thank you.
module 'tensorflow' has no attribute 'Tensor'
What does this even mean?
Since I have no idea myself, this is what Claude gave me:
This error typically means there's a version compatibility issue with TensorFlow. Here's what's likely happening:
The Problem
The code is trying to use tensorflow.Tensor (or tf.Tensor), but in TensorFlow 2.x, tensors are accessed differently than in TensorFlow 1.x. The attribute path has changed between versions.
Common Causes
TensorFlow 2.x compatibility - The workflow or custom nodes might be written for TensorFlow 1.x but you have TensorFlow 2.x installed (or vice versa)
Incorrect import pattern - The code might be using an outdated way to reference tensors
Lazy loading issues - In some TensorFlow 2.x versions, certain modules need to be explicitly imported
Solutions to Try
Check your TensorFlow version:
pip show tensorflowIf using TensorFlow 2.x, the correct way to reference tensors is usually just using the tensor objects directly, or:
tf.constant(), tf.Variable(), etc. return tensor objects
Type checking: use isinstance(x, tf.Tensor) or check the TensorFlow types module
Try reinstalling TensorFlow:
pip uninstall tensorflow pip install tensorflow==2.10.0 # or whatever version Wan Video 2.2 requiresCheck the Wan Video 2.2 requirements - The model likely has specific TensorFlow version requirements in its documentation or requirements.txt file.
Does the error show which file or node is causing this? That would help narrow down whether it's a core dependency issue or a specific custom node problem.
What's the best way to get rid of the slight flicker whenever it loops?
Or perhaps I'm doin' somethin' wrong? Every image I generate with this, and other loops, have a lighting shift when it loops. Let me know if you have any tips or advice, thanks :)
Is the clip you're generating more than 5 seconds long? I noticed a color shift when going over 81 frames. I have a color correct node in the workflow if I remember correctly. Try enabling it or if it is enabled, try to disable it.
Hm, alright at 64 frames and using an image that's clearly stationary (person sitting and writing) there's just barely the slightest dim flicker. I just wonder how some people can get perfect loops without such flickering.
If I use any sort of walking character it won't even loop it back to the frame, it just gray screens and restarts when it gets to the end xD
Slowly getting hang of it, thanks for the tips. I've changed to your all-in-one workflow and it's very good; appreciate the work.
@Vohsy You could also try to reduce the number of frames to 48 (3 seconds) and check if the color shift still occurs. When doing a loop even the slightest mismatch will be visible. Sometimes it's just not working.
Alright, so by stitching the frames with a massive workflow called VACE Clip Joiner v2.4 I've managed to get no flicker at all. Thanks again for the workflow, and thanks for the tip of shorter videos. Appreciate it.
@Vohsy hello,i am using vace clip joiner light,can u tell me what u did to get loop?,u genrated two short loop videos and stitched them {both videos with same first and last frame}?
I just want to say thank you for this. Your videos have inspired me to give video making a try. I was interested in making videos for a while, but I couldn't get anything working with a basic workflow. Also I had trouble finding information for fixing the initial issues I was having. This is the only workflow i have used that has actually given me some somewhat consistent results. I am still learning but I feel like I am actually beginning to take some steps with this workflow. So again, Thank you very much!
Thank you for the comment. That's wonderful to hear you're getting the results you wanted. There is a more recent release of my workflows combining all three into one. Worth checking out. https://civitai.com/models/2404513/i2v-first-last-frame-loop-upscale-and-interpolate-or-all-in-one-wan-22-workflow