🌸 OVERTLI Pollinations Image Generator

Most nodes in the Overtli Studio Suite rewrite text. GZ_ImageGen is one of the ones that makes pictures. It's a cloud image generator: your prompt goes to Pollinations' image API, the bytes come back, and they land on an IMAGE socket that looks and behaves exactly like the output of a local sampler. Same downstream nodes, same preview, same everything - the only difference is there's no checkpoint, no VRAM, and no model download anywhere in the chain.

That's the whole appeal, honestly. If you're on a machine with no GPU (or not enough of one), or you just want a fast draft without loading a checkpoint, this node gets you from prompt to IMAGE tensor in a call to gen.pollinations.ai instead of a diffusion pass. The suite's model dropdown is populated live from Pollinations' catalog - 160+ models tagged with capability markers like [image-gen], [reference-image?], [free]/[paid] - so you're picking from whatever the provider actually offers today, not a stale hardcoded list.

The inputs that matter

• prompt - what you want to see. Multiline.
• model - default is an OpenAI image-gen model tagged [free]. Free-tier models are the reason this node is so frictionless; paid ones need the gems/pollen economy on enter.pollinations.ai, so don't pick a [paid] entry on an empty account.
• width and height - 256–2048, default 1024². Step in multiples of 64.
• image - optional reference image. The [reference-image?] tag on so many models means several can take an input image for editing or style reference, and enhance_prompt (off by default) lets a model rewrite your prompt first.
• mode_preset - a big dropdown mixing text and image modes (Enhance, Tags, Portrait Photography, Concept Art…). It shapes the prompt before generation.
• style_preset_1..3 and additional_styles - the suite's curated style layers, chainable from GZ_StyleStackNode.
• seed - set a fixed value for reproducible-ish drafts (default -1 = random).
• safe_mode and no_logo - content moderation and the Pollinations logo watermark. no_logo defaults to true.

Output

One image socket of type IMAGE - wire it to a preview node, a save node, or into img2img-style downstream work. It's a real tensor once it's in your graph, so you can pass it straight into an upscaler or a second generation step.

Install and honest expectations

Search "Overtli" in ComfyUI Manager, or git clone https://github.com/OvertliDS/overtli-studio-suite into custom_nodes, then pip install -r requirements.txt and restart. Dependencies are featherweight (requests, Pillow, numpy) - again, no model files, because the model is someone else's.

Set your expectations accordingly. Pollinations is a free hosted tier, and community complaints about it cluster around one theme: it's free and generous, but it's also slow, occasionally down, and sometimes inconsistent during spikes. A timeout on a draft is the node working as designed, not a bug. The other reality is that cloud image models have their own aesthetic and their own moderation - you can't reach into this node and patch a refusal, and for NSFW or anything a hosted filter won't serve, this is the wrong tool full stop. Use it for drafts, ideation, and no-GPU workflows, not as the core of a careful local pipeline - and keep an eye on GZ_POLLINATIONS_IMAGE_TIMEOUT if you're doing big batches.