Been trying to build a pipeline: use Qwen’s multi-angle generation to get a character’s front/side/back views out first, keeping the look consistent, then feed those into Krea 2 for a second pass of texture and aesthetic refinement — in theory you’d get both consistency and better-looking output.
Tried stitching a few versions together myself, but got stuck at the handoff — feeding the multi-view outputs into Krea 2 for re-rendering tends to wash out the character’s features, and consistency falls apart.
Anyone actually gotten this chain working? Are you locking the shape with ControlNet in between, or something else? Or maybe these two just shouldn’t be forced together and there’s a smoother combo — happy to hear suggestions.