👑 God Mode (Metadata Fixed)

If you came here because ComfyUI's blank canvas gives you a small panic attack, this is the node you're actually looking for. God Mode collapses the entire standard txt2img/img2img pipeline - checkpoint, VAE, CLIP skip, prompt, five LoRAs, sampler, ControlNet, hires fix, post-upscale, save - into one node with a stack of widgets. It's the A1111 web form rebuilt as a single ComfyUI custom node. The name is a bit much, but the thing itself is honest: type a prompt, click Queue, get an image, done.

Why it exists

ComfyUI's whole pitch is that the graph shows you every step; its whole curse is that you have to build every step. There's a whole genre of "why did you switch from A1111 to ComfyUI" threads where the answer is always the same - once you need something the form UI can't express, there's no going back. God Mode is for the person who wants to stay in ComfyUI (the models, the workflows, the eventual control) but wishes it felt like A1111 for day-to-day generation. It sits in the same niche as Efficiency Nodes, just taken to its logical extreme: one node instead of a dozen, at the price of one very tall node.

How it works

There's no magic here, and that's the good news. Read the source and you'll find it's a thin script that calls ComfyUI's own core nodes in sequence - CheckpointLoaderSimple, LoraLoader, CLIPTextEncode, KSampler, VAEDecode, SaveImage, the lot. That means zero dependencies (the pack's pyproject.toml declares an empty list) and no models to download - every dropdown is just your existing checkpoints/, loras/, controlnet/, and upscale_models/ folders. Whatever you already have in ComfyUI shows up here.

It also does the smart-mode trick: connect an image to input_image_i2i and it flips from text-to-image to img2img, encoding your image and taking its dimensions as the working resolution. Connect one to input_image_cn, pick a ControlNet model, and that condition gets applied to both positive and negative conditioning via ControlNetApplyAdvanced.

The inputs that matter

It has ~30 widgets, but you'll actually touch a handful:

• positive_prompt / negative_prompt - pre-filled with an anime default (masterpiece, best quality, 1girl, scenery). Replace them, obviously.
• ckpt_name + vae_name - the VAE defaults to Baked-VAE, i.e. use the one inside the checkpoint.
• The five LoRA slots (lora_1_name…lora_5_str) - a LoRA stack with independent strengths; setting a strength to 0 skips that slot entirely.
• seed, steps, cfg, sampler_name, scheduler, denoise - the standard sampler block. The sampler list comes straight from ComfyUI's KSampler, so all the usual suspects are there.
• hires_enable, hires_scale, hires_denoise, hires_steps - the hires-fix second pass (latent upscale, then sample again at the same seed).
• post_upscale_model, post_upscale_scale - a model-based upscaler like 4x-UltraSharp on the finished image.

Outputs are IMAGE, MODEL, and VAE, and since it's an output node, the image is saved to ComfyUI/output/ under the default prefix GodMode/Image.

Install

Two ways, both trivial - the README says so and the source agrees (no requirements.txt, no pip install):

cd ComfyUI/custom_nodes
git clone https://github.com/g7b2/ComfyUI-GodMode-Node.git

Restart ComfyUI, then right-click → Custom OneClick → 👑 God Mode (Metadata Fixed). Or install via ComfyUI Manager → "Install via Git URL" with the same repo URL.

Where people get burned

• Width/height are ignored in img2img mode. The node takes its resolution from your input image, not those widgets - your denoise is what controls how much it changes. Don't set 1024×1024 expecting an upscale.
• CLIP skip does nothing at -1, which is the default; only values below that (like -2) actually engage, and only on models that expect it - the SD 1.5 / anime lineage per the CLIP-skip lore.
• The display name says "(Metadata Fixed)" for a reason. An earlier version dropped the workflow metadata; this one passes the prompt and PNG info through to SaveImage, so dragging an output PNG back onto the canvas reconstructs every setting. That's the standard "the image is the project file" convention - keep the original PNG, since any re-encode strips it.

The honest verdict: this is a tiny pack - one commit, one node, almost no community footprint - and it's a convenience ramp, not a career workflow. Like any custom node it runs arbitrary Python, so the usual "read the __init__.py before you trust it" rule applies; this one is ~300 lines, so that's cheap. Use it to stop fighting the graph and start making images, then unwire it into the real thing when you outgrow it - which is exactly what it's for.