Flux Context: Image-to-Panorama Workflow in ComfyUI

Just landed a gig doing an indoor display, and the client wanted those wide panoramic shots you can swipe left and right on. Perfect timing to try out Flux Kontext’s image-to-panorama workflow. The trick isn’t just stretching or cropping—it lets the model fill in content based on your original image’s perspective and lighting, expanding horizontally up to three times the width. I hit a snag myself: never feed it an input image that already has 3D perspective or an isometric view, or the panorama will warp and the seams won’t line up. Stick with a straight-on, eye-level shot with decent resolution, and you’ll get the most stable results.

You’ll need eight models ready—flux1-kontext-dev, t5xxl_fp16, ae are the main ones, plus an HDR360 LoRA and a 4x upscaler. For the upscaling step, only set it to x2; I got greedy and cranked it higher, and it crashed hard. The seam quality after running it is super smooth—perfect for VR environments or immersive displays, no obvious fake breaks. VRAM usage is brutal; 24GB barely handles it, but for peace of mind, go with 48GB.

That “front-facing eye-level” tip is so real. I tried feeding it perspective shots before and they all came out twisted like pretzels.

Totally feel you on the x2 cap thing—one more step and my VRAM would’ve blown up.

Honestly, can you really make the seams invisible for indoor shots?

Yeah, I’ve been testing it too—windows and corners are the easiest places to mess up. Gotta rely on the prompt to patch things up.

Is that HDR360 LoRA a must-have? Does it make a big difference if you don’t add it?

48G just to be safe? Yeah, that’s a pretty high bar, definitely scared off a bunch of people.

Bookmarked this, gonna try it out this weekend with some real construction site photos.