Anima Turbo · Edit

Anima Turbo Edit is the odd one out in this pack: not ByteDance, not four credits, not a 2K tier. It's Circlestone Labs' Anima - the anime model that made people switch off the Illustrious/NoobAI stack in 2026 - in its distilled turbo form, as an image-to-image node, for about 1.8 credits an image. Cheapest thing in the pack by a wide margin, which makes it the safest place to learn how these nodes behave.

You feed it a picture and a description of the picture you want back, and it restyles toward it. Photo to anime, sketch to finished frame, rough anime draft to a cleaner one.

Why Anima specifically

Anima isn't SDXL with anime weights. It's a 2B DiT built on NVIDIA's Cosmos-Predict2, with a Qwen3-0.6B text encoder and the Qwen-Image VAE. The consequence that matters day to day is prompt comprehension: it was never trained into Danbooru tags the way Illustrious and NoobAI were, so plain natural language works - and its grip on general world knowledge survived. Ask for a specific pose, camera angle or number of characters in sentences and it tends to follow.

Locally, Anima is known for a low VRAM floor and cheap training; this node doesn't care about either, because the generation happens on PopNow's servers. What you're buying is the model's look without the 6GB-plus download and the multi-step sampling - Turbo is the distilled few-step variant, which is why the price is where it is.

The inputs

• prompt - and here the tooltip is unusually good advice: describe the image you want back, not just the change. "Change the background to a rooftop at dusk" is a weaker prompt than "an anime girl in a school uniform standing on a rooftop at dusk, warm rim light, city below". The model is restyling a whole frame; talk about the whole frame.
• image_url - the source IMAGE. The result keeps its aspect ratio - there's no ratio widget on this node, so crop your input first if you need a specific shape.
• resolution - 1k (about one megapixel, the default) or 1.5k (about 2.4). Both keep the source's ratio. 1k is fine for iteration; 1.5k for anything you'll keep.
• strength - 0 to 1, default 0.8, and this is the dial to actually learn. Note what the tooltip says it means: how far to move away from the source. At 0.8 the output is mostly the model's invention, guided by your picture. Drop to 0.35–0.5 for a photo-to-anime conversion that still looks like the person; push to 0.9+ for a loose reinterpretation.
• seed - 0 to 2147483647, and the only node in this pack's image section that gives you one. Useful for controlled comparison: same seed, strength 0.5 versus 0.7, see exactly what moves.

Outputs

images (a batch) and task id. No alpha, no mask input - this is a whole-frame restyle, not inpainting. If you want to change one region, mask it locally and composite afterwards.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/popnow-ai/popnow-comfyui

Or Manager → PopNow → Install. Key from popnow.ai/api/console/keys into Settings > PopNow or POPNOW_API_KEY, then restart again so the catalog fills the node list. The pack has no pip dependencies at all; Python 3.10+ and a stock ComfyUI is the entire requirement.

Common issues

• The result doesn't look like the person - strength too high. 0.8 is a lot. Try 0.45 and climb.
• The result looks like the photo with a filter - strength too low, or the prompt described the source rather than the target.
• Anime style bleeding into a photo you wanted kept - that's the model doing its job. This is an anime model; if you want a photo restyle, a different model in this pack is the right call.
• PopNow 40201 - credits. At ~1.8 per image it's hard to run dry by accident, unlike the video nodes.
• PopNow 40004 - unsupported combination; usually an extreme input size against the resolution tier.
• Cancelling stops your wait, not the job - it finishes server-side and bills on success.

At under two credits a frame this is the node I'd point a beginner at first, not because it's representative of the pack - it isn't; most of this thing costs real money per second - but because it's a cheap, fast way to feel out how hosted ComfyUI nodes behave before you point a Seedance node at your card.