Anima Turbo LoRA · Text to Image

Anime without the checkpoint download

Anima is one of the few genuinely interesting things to happen to anime generation since Illustrious - a 2B base from CircleStone Labs sitting on Cosmos-Predict2 instead of the SDXL stack everyone else kept iterating. Turbo is the distilled version, and distillation always trades a little quality for a lot of speed: the whole point of the technique is collapsing a twenty-step trajectory into one or two hops. The community proof is mundane and convincing - people report an image going from "minutes" to under ninety seconds on a weak card with the Turbo-LoRA attached.

This node is that, hosted. You get circlestone-labs/anima-turbo-lora/text-to-image with your own LoRAs stacked on top, and the only thing your machine does is send a prompt and receive a PNG.

How it runs

Nothing is sampled locally. The node posts your prompt and the resolved LoRA references to PopNow, gets a price quote, submits the task under an idempotency key, then polls it to completion and decodes the returned files into an IMAGE batch. The LoRAs themselves come in through the loras socket, fed by PopNow LoRA nodes - a local .safetensors in ComfyUI/models/loras gets uploaded once and reused for twelve hours; a Hugging Face repo id, Civitai page, or direct link gets imported server-side.

Moderation is the vendor's, not yours. There is no abliteration trick here: the filter lives behind the endpoint, and everything you type leaves your machine. If the base model's licence and unrestrained weights were the reason you picked Anima in the first place, understand that this particular door is a service with an acceptable-use policy attached.

Inputs that actually change the picture

• prompt - Danbooru-style tags, plain English, or a mix; the model takes both. The tooltip's advice matters: put LoRA trigger words first. Tag order still carries weight.
• aspect_ratio - default 1:1. This model wants a clean ratio; don't fight it.
• resolution - 1k (roughly a megapixel) or 1.5k (roughly 2.4). Same composition either way, more pixels to work with later.
• seed - starts on fixed. Two identical queues produce one billed image and one cached result. Set control-after-generate to randomize when you're exploring, and expect to pay for each roll.
• loras (optional) - up to three Anima-compatible weight files, each with its own strength on the PopNow LoRA node. This is the whole reason to use this variant instead of plain Anima Turbo.

Outputs: images (IMAGE) and task id (STRING, useful for the API console logs or a support ticket).

Install and key

Manager → search PopNow → Install, or from a shell:

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

Restart ComfyUI, grab a key from popnow.ai/api/console/keys, paste it under Settings > PopNow (or export POPNOW_API_KEY for a server), restart again so the catalog loads. There are no model downloads, no Python extras beyond what ComfyUI ships, and no GPU requirement - but you do want a recent ComfyUI, since the pack is built on the V3 node API (comfy_api.latest).

Practical notes

The pricing is the friendly part: from about 2.7 credits per image, quoted before every run, with failed runs refunded. Set Max credits per run in the PopNow settings if you're the sort of person who leaves a randomize seed running overnight.

Where it bites: PopNow 401 is a key problem, 40201 is an empty wallet, and 40004 means the field combination isn't servable - read the named field rather than re-queueing hopefully. Feeding it a LoRA trained for a different architecture gets you nothing useful; the socket is labelled Anima for a reason. And cancelling only stops the wait, not the job - the task keeps running on PopNow and is charged if it lands.

One honest caveat before you build a workflow around it: PopNow is new and has essentially no community footprint yet, so you're an early adopter on a vendor API rather than a user of a pack with a thousand threads of accumulated folklore. Read what the node sends, keep the key in the environment, and start with a small credit pack.