MemeplexRender

MemeplexRender is the workhorse of the shi3z/ComfyUI_Memeplex_DALLE pack and the node most people land on this pack for. It sends your prompt to memeplex.app, a paid cloud service that generates anime and illustration images with its own fine-tuned models, and pulls the finished pictures back into ComfyUI as a normal IMAGE batch. Zero VRAM, zero checkpoint downloads - your only hardware requirement is an internet connection and an account with memeplex.app.

Reach for it when you want consistent anime output but don't want to babysit an SDXL anime fine-tune, or when your machine simply can't run one. The tradeoff is the same as with any cloud renderer: you pay per image, and the queue is someone else's server load. If you're already running Illustrious or NoobAI locally, this node won't beat them on quality or control - it's the convenience option.

How it works

Straightforward under the hood. The node builds a GET request to https://memeplex.app/api/ with your account's uid and apikey in the query string, plus the prompt, size, quantity, and model. The sampling steps are hardcoded to 30 in the source. The API returns URLs; the node then polls each one every 30 seconds until the image is ready, converts it to a float32 tensor, and stacks the whole batch together into a single IMAGE output.

Inputs that matter

• prompt (required) - the main event, a forced string input. Wire it from a text or GPT node.
• qty - how many images per run, 1–9, default 9. Each one costs you a credit, so don't leave it at 9 if you're just iterating on a prompt.
• width / height - 512–2048, snapped to steps of 64, default 1024 each.
• model - one of SDXL1.0, trinart, StableDiffusion-v1-5, StableDiffusion-v2-0. SDXL1.0 is the sensible default for anime.
• negative (optional) - a forced string input; leave it unwired unless you have a specific anti-list.

The IMAGE output is a batch of qty frames, so PreviewImage will show them as a grid and SaveImage writes them out as a folder. That's the natural way to see everything at once.

Installing it

ComfyUI Manager can find it by searching the pack title, or clone it by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/shi3z/ComfyUI_Memeplex_DALLE
pip install openai

There's no requirements.txt, so the openai package won't be auto-installed - and the whole pack refuses to load without it, even though MemeplexRender itself never talks to OpenAI. The real setup step, per the README's terse instruction, is editing nodes.py to fill in the uid and apikey variables from your memeplex.app account. Those are the service's keys, not OpenAI's. Restart ComfyUI after editing.

Where people get burned

The most common failure looks like the node hanging forever on "request queing": that's usually a wrong or missing memeplex key - the code prints the API response and silently retries. Be patient with legitimately slow runs too; cloud queues can take a minute or two, and the node blocks the whole graph while it polls. One quirk worth knowing: the prompt is shoved straight into the URL query string without encoding, so prompts with spaces or odd characters can break the request - which is exactly why the pack's GPT node is built to force a clean, all-English prompt. Feed this node tidy prompts and it's remarkably painless for a paid service wrapper.