Better character face fidelity
Hi. I'm trying to get the best fidelity for my character face. It's good with MSR but not as good as when I insert the character picture in the background with a classical I2V workflow. What can I do?
-Upscale the face reference image resolution?
-Switch from LTX distill to DEV model version?
-Train a character Lora and add it in the workflow ?
-Is there any parameter that could be tweaked or a node to add that could upgrade the quality?
Thanks for your help.
Hi. I'm trying to get the best fidelity for my character face. It's good with MSR but not as good as when I insert the character picture in the background with a classical I2V workflow. What can I do?
-Upscale the face reference image resolution?
-Switch from LTX distill to DEV model version?
-Train a character Lora and add it in the workflow ?
-Is there any parameter that could be tweaked or a node to add that could upgrade the quality?
Thanks for your help.
Emmm… my answer would probably be: wait for the next version.
The reason is that MSR works differently from a classical I2V workflow. MSR is closer to T2V, so it needs to balance the model’s attention between the reference image and the text prompt.
For now, maybe you can try simplifying the description of the reference image as much as possible. If the prompt describes the reference too heavily, the model may rely more on the text than the actual visual reference, which can reduce facial fidelity.
Thank you very much !
When do you think to have a new version ?