Ryder

If you've ever run Pony Diffusion V6, you know the ritual: every prompt has to open with score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, highly detailed before any actual content. The model learned that whole concatenated string as one quality trigger during training, so it's effectively mandatory. EasyPony (displayed in ComfyUI as "Ryder") turns that ritual into one dropdown. It's two nodes fused into one: it loads a Pony checkpoint and writes your positive and negative prompts, with the score/source/rating conventions baked in.

Worth hearing the author's framing before you install: they built it as a helper for one of their other (unpublished) nodes, they're "not a prolific user of Pony models," and they say YMMV on prompt quality. It's the distilled implementation of four CivitAI articles about Pony prompting, including one by the model's creator. So yes, it's a prompt generator - but there's method to the madness.

What it actually outputs

Five outputs: MODEL, VAE, CLIP, PROMPT, and NEGATIVE. That means it replaces your Load Checkpoint node and the typing you'd normally do into two CLIP Text Encode nodes. The stock workflow wires CLIP + PROMPT into one CLIP Text Encode, CLIP + NEGATIVE into the other, MODEL into the KSampler, and VAE into VAE Decode. So it saves you a node and a lot of copy-pasting, but you still need the two Text Encode nodes to turn the strings into conditioning.

The inputs that matter

• Quality - the one you'll actually touch. EVERYTHING (default) emits the full official score_9 through score_4_up opener; GOOD is score_9, score_8_up, score_7_up; ONLY_THE_BEST is just score_9; AVERAGE stops at score_5_up. It's a dial for how picky the model is about what enters the frame.
• Source and Rating - the source_anime/pony/fury/cartoon and rating_safe/questionable/explicit filters. Leave at "-" to skip.
• Invert Source (Neg) / Invert Rating (Neg) - push the chosen tag into the negative prompt instead, so you can steer away from a style or rating rather than toward it.
• SFW - on by default; adds (sfw:1.2) to the positive and a censorship string to the negative.
• Quality Boost (Beta) and Negative Boost (Beta) - bolt large negative keyword lists onto the negative. Beta for a reason; more below.
• Stop at Clip Layer - defaults to -2, which is CLIP skip 2, the setting every Pony guide tells you to use. You basically never touch it; setting it to 1 is what turns Pony output into noise.
• prefix / suffix (optional) - prefix replaces the quality score entirely (handy if your Pony derivative wants a custom score string), suffix just appends.

Then Prompt is the only text you write: your booru tags. And everything gets lowercased on the way out, so don't bother with capital letters.

Install

Search "Easy Pony" in ComfyUI Manager, or:

cd ComfyUI/custom_nodes
git clone https://github.com/regiellis/ComfyUI-EasyPony

then restart ComfyUI. There's no dependency list and nothing to download - requirements.txt is empty, and the node just loads whatever Pony checkpoint you already have in your checkpoints folder. One wrinkle: the README's manual clone command still points at itsjustregi/ComfyUI-EasyPony, but the repo lives under regiellis. Manager is the safe route; if you clone by hand, use the URL above.

Common issues

• Only use it with Pony-family checkpoints. The tooltip says it plainly: "ONLY USE PONY MODELS HERE." The score/source/rating tags are inert on Illustrious, NoobAI, base SDXL, and Flux - you'd just be pasting a magic string that does nothing.
• It's built on V6 conventions. Pony V7 kept the vocabulary, but its own model card admits the quality tags "have much weaker performance compared to V6." On V7, this node is mostly a no-op past the loading.
• The Boost toggles dump boilerplate. The lists are the SD 1.5-era kitchen-sink negatives - bad anatomy, watermark, even bad parenting. Some of it helps, most of it is noise that the community has been arguing about since 2022. They're off by default for a reason.
• Update gotcha. As of v1.2.0 the node gained the Model dropdown and invert toggles, and the author warns old flows need rework - you may need to delete and re-add the node after updating.

Set Quality to GOOD or EVERYTHING on a first run, leave the betas off, and you have a one-node Pony pipeline that just works.