Nodes/ComfyUI_Memeplex_DALLE/MemeplexCustomSDXLRender
ComfyUI Node

MemeplexCustomSDXLRender

One model, no options, and four images you didn't configure

By shi3z·Created 2 years ago·Updated 2 years ago· 2
MemeplexCustomSDXLRender
    • IMAGE
    prompt
    negative

    MemeplexCustomSDXLRender is the odd one out in shi3z/ComfyUI_Memeplex_DALLE, and it reads like it was built for the author's own use. Its sibling MemeplexRender gives you width, height, quantity, and a model dropdown; this node gives you a prompt box, an optional negative box, and nothing else. Under the hood it always calls memeplex.app with the same fixed config - model custom_sdxl_anime2, 1024×1024, 30 steps - and then, here's the strange part, it does string surgery on the returned image URL to fetch three extra frames and stacks everything into a 4-image IMAGE batch.

    I'll be honest about what that URL manipulation looks like from the source: it takes the API's URL and does url.replace("0_", "1_0_"), "2_1_0_", "3_2_1_0_" to build three more URLs, then downloads all four. That smells like reaching into the API's internal per-step naming scheme to grab intermediate frames rather than any documented feature. It works, and you get four images per run, but you're depending on the URL format never changing.

    When you'd use it

    Honestly? Only if you specifically want what custom_sdxl_anime2 outputs and you don't want to think about settings. It's the "one button" variant - wire a text node in, queue, wait, get a grid. If you want any control at all - different model, different size, a specific batch count - use MemeplexRender instead. This node is the equivalent of a camera with the lens glued on.

    Inputs and output

    • prompt (required) - a forced string input.
    • negative (optional) - a forced string input, leave unwired if you don't need it.
    • Output: IMAGE, a batch of four images.

    Because the inputs are forced strings, they expect to be wired from text or GPT nodes rather than typed in - another sign this was built for the author's own pipeline.

    Installing and running it

    Identical to the rest of the pack: clone into custom_nodes, pip install openai (the pack won't load without it), and edit nodes.py to fill in your memeplex.app uid and apikey - the README's "fill API keys in code" instruction is literal, there's no settings UI. Restart ComfyUI afterward.

    Gotchas

    Beyond the fixed config, the thing to watch is that four-image batch. It's a side effect of guessing at the API's URL scheme, so if memeplex.app ever changes its URL format, this node breaks while MemeplexRender keeps working - the 30-second retry loop will just spin forever on URLs that 404. If that happens, your move is to switch to MemeplexRender, which uses the documented request path. Treat this node as the quirky, convenient, slightly fragile option it is.

    CategoryMemeplex

    Inputs (2)

    NameTypeDefaultDescription
    promptSTRING
    negativeoptSTRING

    Outputs (1)

    NameTypeDescription
    IMAGEIMAGE