vybs-ai/vybs-loras

VYBS LoRAs

Hosting repo for VYBS style LoRAs, published public so they can be loaded by bare URL from fal.ai workflows. fal fetches loras[].path without any Hugging Face token, so a private repo would fail with 401 partway through a generation.

---

URL pattern

https://huggingface.co/vybs-ai/vybs-loras/resolve/main/<path-in-repo>

Use /resolve/main/. Do not use /blob/main/ — that serves an HTML preview page, and fal will fail to parse it as a tensor file.

✅	https://huggingface.co/vybs-ai/vybs-loras/resolve/main/2d/flux/vybs-2d-flux-dev-v1.safetensors
❌	https://huggingface.co/vybs-ai/vybs-loras/blob/main/2d/flux/vybs-2d-flux-dev-v1.safetensors

---

Folder layout

Folders are organised base model first. This is load-bearing, not cosmetic: a Flux LoRA will not load on WAN, Qwen-Image or SDXL. The folder tells you which base a file is loadable against before you ever open it.

2d/
  flux/       Flux (dev / turbo)     — transformer.single_transformer_blocks.*
  qwen/       Qwen-Image             — modelspec.architecture = qi/lora
  krea2/      Krea 2 (fal/krea-2)    — blocks.*.attn.wk
  ideogram/   Ideogram v4            — conditional_transformer.layers.*
  z-image/    Z-Image Turbo          — transformer.layers.*.adaLN_modulation
wan/          WAN 2.2                — transformer.blocks.*.attn1.*
characters/   reserved — character LoRAs (empty)
products/     reserved — product LoRAs (empty)

characters/ and products/ are reserved for future non-style LoRAs. There is currently no SDXL LoRA in this repo, so there is no 2d/sdxl/ folder; add one when the first SDXL run lands.

---

Naming convention

vybs-<domain>-<style>-v<N>.safetensors

• lowercase, hyphens only, no spaces or underscores
• <domain> — content domain (2d for the brand illustration style)
• <style> — the base-model flavour the LoRA was trained against
• v<N> — explicit version, bumped on every retrain; never overwrite a version

The base model is repeated in the filename on purpose, so a file that gets downloaded out of its folder is still self-describing and can't be loaded onto the wrong base.

---

Contents

Repo path	Base model	Trigger phrase	Steps	Size
2d/flux/vybs-2d-flux-dev-v1.safetensors	Flux dev	(none recorded)	1000	124.9 MB
2d/flux/vybs-2d-flux-turbo-v1.safetensors	Flux turbo	ohwx-Vybs-turbo-flux-trainer	1000	123.2 MB
2d/qwen/vybs-2d-qwen-image-v1.safetensors	Qwen-Image	(none recorded)	1000	562.8 MB
2d/krea2/vybs-2d-krea2-v1.safetensors	Krea 2 (fal/krea-2)	vybs-illustration-Krea2	100	223.8 MB
2d/ideogram/vybs-2d-ideogram-v4-v1.safetensors	Ideogram v4	vybs-ideogram	1000	81.4 MB
2d/z-image/vybs-2d-z-image-turbo-v1.safetensors	Z-Image Turbo	Vybs-z-image-turbo-trainer-v2	2000	81.2 MB
wan/vybs-2d-wan-22-a-v1.safetensors	WAN 2.2	vybs-art-wan-22	1000	292.6 MB
wan/vybs-2d-wan-22-b-v1.safetensors	WAN 2.2	vybs-art-wan-22	1000	292.6 MB

Qwen-Image LoRA: rank 32, alpha 1. Krea 2 LoRA: rank 32, alpha 32, bf16.

⚠️ The two WAN 2.2 adapters

WAN 2.2 A14B uses a high-noise / low-noise expert pair, and these two files are almost certainly that pair — identical size and tensor layout, emitted two seconds apart by the same training job. Neither file records which expert it is, so they are named neutrally rather than guessed at:

File	SHA-256
wan/vybs-2d-wan-22-a-v1.safetensors	1b7241d315c7256e3792969f6aa5a275c18917e972fd759458d5b2129cdcaa34
wan/vybs-2d-wan-22-b-v1.safetensors	5c77561513005fc9d7cc362a3f88ccdb662ede775ba1168eefecea52e7d1316b

Once the mapping is confirmed, rename to -high-noise-v1 / -low-noise-v1.

---

Using these from fal.ai

fal-ai/flux-lora accepts a bare URL in loras[].path — no auth header, which is exactly why this repo is public.

These examples only apply to the two Flux LoRAs. The Qwen-Image, Krea 2, Ideogram v4, Z-Image and WAN files each need their own base-matched fal endpoint; passing them to fal-ai/flux-lora will fail to load. Check fal's model catalogue for the current endpoint id per base.

JSON

{
  "prompt": "a toucan perched on a branch, vybs 2d illustration style",
  "image_size": "square_hd",
  "num_inference_steps": 28,
  "guidance_scale": 3.5,
  "num_images": 1,
  "loras": [
    {
      "path": "https://huggingface.co/vybs-ai/vybs-loras/resolve/main/2d/flux/vybs-2d-flux-dev-v1.safetensors",
      "scale": 1.0
    }
  ]
}

JavaScript — @fal-ai/client

import { fal } from "@fal-ai/client";

fal.config({ credentials: process.env.FAL_KEY });

const result = await fal.subscribe("fal-ai/flux-lora", {
  input: {
    prompt: "a toucan perched on a branch, vybs 2d illustration style",
    image_size: "square_hd",
    num_inference_steps: 28,
    guidance_scale: 3.5,
    loras: [
      {
        path: "https://huggingface.co/vybs-ai/vybs-loras/resolve/main/2d/flux/vybs-2d-flux-dev-v1.safetensors",
        scale: 1.0,
      },
    ],
  },
  logs: true,
  onQueueUpdate: (update) => {
    if (update.status === "IN_PROGRESS") {
      update.logs.map((l) => l.message).forEach(console.log);
    }
  },
});

console.log(result.data.images[0].url);

Install with npm install @fal-ai/client.

Python — fal_client

import os
import fal_client

os.environ["FAL_KEY"] = "..."  # or export it in your shell

LORA_URL = (
    "https://huggingface.co/vybs-ai/vybs-loras/resolve/main/"
    "2d/flux/vybs-2d-flux-dev-v1.safetensors"
)

def on_queue_update(update):
    if isinstance(update, fal_client.InProgress):
        for log in update.logs:
            print(log["message"])

result = fal_client.subscribe(
    "fal-ai/flux-lora",
    arguments={
        "prompt": "a toucan perched on a branch, vybs 2d illustration style",
        "image_size": "square_hd",
        "num_inference_steps": 28,
        "guidance_scale": 3.5,
        "loras": [{"path": LORA_URL, "scale": 1.0}],
    },
    with_logs=True,
    on_queue_update=on_queue_update,
)

print(result["images"][0]["url"])

Install with pip install fal-client.

---

Adding a new LoRA

hf upload vybs-ai/vybs-loras <local-file> <base-folder>/vybs-<domain>-<style>-v<N>.safetensors --no-private

Then confirm the URL serves the file rather than an HTML page:

curl -sIL "https://huggingface.co/vybs-ai/vybs-loras/resolve/main/<path>" | grep -iE "^HTTP|^content-type|^location"

Expect a final HTTP/2 200 with content-type: application/octet-stream (via a location: redirect to the CDN). A content-type: text/html means the path is wrong or you used /blob/main/.