Dream Panorama Viewer Pro

Let's get the first lie out of the way: DreamViewer doesn't interpret anything. Despite the pack's name, this node is a viewer - the payoff at the end of the Dream-Interpreter graph. You hand it a 2:1 equirectangular panorama plus a text interpretation, and it drops the image into an interactive 360° canvas you can orbit, zoom, tilt, auto-rotate, and screenshot. If you've ever generated a panorama and squinted at the seam in a flat image editor, this is the node that finally lets you stand inside it.

It's the terminal node of the pack, the one that makes the whole thing feel like a product instead of a pile of math. The interpretation text is shown in a side panel next to the viewer, so the graph reads like a tiny "tell me your dream" app.

How it works

Under the hood it's an output node that does a small, useful job: on every run it saves your hdri_image as a PNG into ComfyUI's temp directory, then the pack's JavaScript extension (a bundled, locally-pinned Three.js - no CDN, no telemetry) projects it onto a 360° sphere in your browser. The controls are the standard orbit/zoom plus adjustable field of view, PNG capture, and reduced-motion-aware auto-rotation.

The clever bit: the viewer runs on your browser's WebGL, not on CUDA. It works the same on a potato laptop as on the RTX 3090 doing the generation. No model weights, no network calls, nothing to download beyond the pack itself.

The two inputs that matter

That's the whole node - two required inputs, no optional ones, so there's nothing to tune:

• hdri_image (IMAGE) - your panorama. It must be a true 2:1 equirectangular image, or the wrap will look like a funhouse mirror. Generate with a panorama LoRA or run it through the pack's Validate and Seam Repair nodes first.
• dream_interpretation (STRING, multiline) - the text that appears in the interpretation panel. Type it, or wire it up from upstream: the original 2024 build of this pack fed it from LLaVA 1.6 Mistral (a GGUF you had to download), but the current v3.0.0 rewrite is a fully offline toolkit where you supply the words yourself. There's a DreamPromptBuilder and DreamMotifTags in the same pack if you want help generating the text.

Install

ComfyUI Manager: search "ComfyUI-Dream-Interpreter" and hit install. Or the manual route:

cd ComfyUI/custom_nodes
git clone https://github.com/gokayfem/ComfyUI-Dream-Interpreter.git
python -m pip install -r ComfyUI-Dream-Interpreter/requirements.txt

Then restart ComfyUI. The dependency list is comically small for a pack this flashy - just numpy and Pillow. No CUDA libs, no transformers. You'll find the node under visualization/3D.

Where people trip up

• Feeding it a normal image. An SDXL square will not wrap into a seamless world. If it looks broken, that's not the node - your input isn't equirectangular. Use the pack's DreamPanoramaValidate to check the geometry and repair the aspect before it hits the viewer.
• Expecting it to interpret. It won't. The interpretation is an input you provide. The old README oversold this; the current one is refreshingly honest that the text utilities make "no psychological or clinical claims."
• The 2024 reconnection bug. The original release required you to delete and re-add the node between runs to refresh the view - a real, author-acknowledged annoyance. v3.0.0 rebuilt the viewer with batch support and cached output restoration, so modern installs shouldn't hit it. If you're loading an old shared workflow and the viewer stalls, this is history biting you.
• Remote setups. The viewer lives in your browser, so if the graph runs on a server you're connected to, the preview still renders client-side. Generation hardware and viewing hardware don't need to be the same machine.

It's a niche utility - you need to be making panoramas for it to earn its keep. But if you are, it's the one you'd actually reach for: zero setup, offline, and it turns a flat seam-ridden image into something you can wander around.