feat: decouple R2V reference images from V2V input video conditioning #302
Conversation
Add vace_use_input_video parameter to separate reference image conditioning from input video conditioning. These are orthogonal concerns that should not be inextricably linked. This will be equally important for FFLF. Signed-off-by: RyanOnTheInside <7623207+ryanontheinside@users.noreply.github.com>
yondonfu
left a comment
yondonfu
left a comment
There was a problem hiding this comment.
Is the main reason for this change that we want to support using VACE ref images while using "normal V2V" eg use input video to generate initial noise?
If so, let's use this opportunity to clarify terminology for both ourselves and the tooltips/descriptions since I think they are a bit confusing right now saying things like "when we are in V2V mode...", "so we can use reference images in V2V mode" - it feels like we're using "V2V" to refer to normal V2V, VACE V2V and when we are using the Video input mode in the UI/API!
I suggest:
- Video input mode in the UI -> "Video input mode"
- VACE V2V -> "VACE conditioning"
- Normal V2V -> "latent initialization"
"In Video input mode we can either use the input video for VACE conditioning or for latent initialization."
And this is how we would talk about these terms in tooltips/descriptions (the current Use Input Video label in the Settings panel feels fine unless there is another concise term that label to mind).
WDYT?
Signed-off-by: RyanOnTheInside <7623207+ryanontheinside@users.noreply.github.com>
ryanontheinside
force-pushed
the
ryanontheinside/feat/separate-vace-conditioning-r2v
branch
from
3f34844 to
568529e
Compare
yondonfu
deleted the
ryanontheinside/feat/separate-vace-conditioning-r2v
branch
Add vace_use_input_video parameter to separate reference image conditioning from input video conditioning. These are orthogonal concerns that should not be inextricably linked. This will be equally important for FFLF.