Fibo 2.0 Psychedelic LoRA

A psychedelic-pop illustration LoRA for BRIA Fibo 2.0 Dev. It emphasizes saturated neon color, surreal subjects, bold black outlines, rhythmic patterns, luminous details, and graphic poster-like forms while preserving Fibo's structured-prompt control.

The adapter was trained with the trigger fibo_psychpop_style. Local evaluation found LoRA scale 0.7 to be the best balance of style strength, composition, and subject fidelity. Scale 1.0 is available for a more forceful graphic treatment.

License: Fibo 2.0 Dev is gated for non-commercial use under BRIA's Fibo terms. The base repository identifies CC BY-NC 4.0 and states that commercial use requires a separate BRIA agreement. Review the base-model license and access terms before use or redistribution.

Files

File	Description
`pytorch_lora_weights.safetensors`	Diffusers / PEFT adapter
`Fibo-2.0-Psychedelic-LoRA-ComfyUI.safetensors`	Native ComfyUI adapter
`adapter_config.json`	Diffusers adapter configuration
`scripts/generate_fibo_lora.py`	Ready-to-run structured-prompt Diffusers example
`images/`	Model-card examples

ComfyUI compatibility

The root pytorch_lora_weights.safetensors uses separate Diffusers / PEFT projections such as transformer.layers.0.attention.to_q.lora_A.weight. ComfyUI's native Fibo 2.0 model uses fused Q/K/V, K/V, and SwiGLU W1/W3 projections, so loading the root file directly can report lora key not loaded or leave important layers unapplied.

Use Fibo-2.0-Psychedelic-LoRA-ComfyUI.safetensors with ComfyUI.

Update ComfyUI to a version with native Fibo 2.0 support; commit 87ed5a0 or newer contains the required model support.
Install the Fibo Dev diffusion model, Qwen3-VL text encoder, and Flux.2 VAE from Comfy-Org/Fibo-2.0 in the folders documented on that model card. Use the Dev model, not Edit Turbo, because this adapter was trained against briaai/Fibo-2.0-dev.
Put Fibo-2.0-Psychedelic-LoRA-ComfyUI.safetensors in ComfyUI/models/loras/.
Use ComfyUI's standard LoRA loader with model strength 0.7 and clip strength 0.0. Scale 1.0 is the tested stronger alternative.
Use Fibo's structured prompt encoder and place fibo_psychpop_style at the beginning of style_medium.

Download only the converted variant with:

hf download axiomofmind/Fibo-2.0-psychedelic-lora \
  Fibo-2.0-Psychedelic-LoRA-ComfyUI.safetensors \
  --local-dir ./ComfyUI/models/loras

The ComfyUI file contains the same adapter as the Diffusers weight in Fibo's fused native layout. The recommended strength remains 0.7.

Recommended settings

Base model: briaai/Fibo-2.0-dev
Trigger: fibo_psychpop_style
Preferred LoRA scale: 0.7
Stronger LoRA scale: 1.0
Inference steps: 50
Guidance scale: 5.0
Resolution: 1024 x 1024
Prompt format: structured Thales VGL JSON

Place the trigger at the beginning of style_medium, for example:

"style_medium": "fibo_psychpop_style, digital illustration"

Structured JSON prompt example

Pass the serialized JSON object as Fibo's positive prompt:

{
  "short_description": "A glowing mushroom with wavy blue and pink caps rises from a starry field against a black background.",
  "objects": [
    {
      "description": "a tall luminous mushroom with layered wavy blue and pink caps",
      "location": "center of the frame"
    }
  ],
  "background_setting": "A black nocturnal field fading into a star-filled void.",
  "lighting": {
    "description": "The mushroom casts blue and pink neon light."
  },
  "aesthetics": {
    "composition": "A centered vertical composition with ample negative space.",
    "color_palette": "electric blue, hot pink, violet, and black",
    "mood": "enchanted and otherworldly"
  },
  "style_medium": "fibo_psychpop_style, digital illustration",
  "context": "A surreal psychedelic night scene.",
  "artistic_style": "luminous contours, fluid shapes, high contrast",
  "text_render": []
}

The structured comparison images used this negative prompt at guidance scale 5.0:

{
  "short_description": "",
  "objects": [
    {
      "description": "the primary subject",
      "texture": "flat, waxy, airbrushed, or lacking fine surface detail",
      "appearance_details": "blurred edges, malformed details, or simplified geometry"
    }
  ],
  "background_setting": "featureless studio gradient with no material detail",
  "lighting": {},
  "aesthetics": {
    "artifacts": [
      "oversaturation",
      "color cast",
      "clipped highlights",
      "lens flare",
      "motion blur",
      "compression artifacts"
    ]
  },
  "photographic_characteristics": {
    "depth_of_field": "excessive bokeh obscuring the scene",
    "focus": "soft or out of focus"
  },
  "style_medium": "",
  "context": "",
  "artistic_style": ""
}

The plain-text comparison examples below instead use an empty negative prompt.

Scale comparison

All columns use the same structured scene prompt, seed, resolution, inference steps, and guidance scale. The stock column disables the adapter and removes the trigger. The LoRA columns enable the adapter and include the trigger.

Prompt	Fibo stock	LoRA 0.7 — preferred	LoRA 1.0 — stronger
0 — Galaxy ram, seed 42
1 — Glowing mushroom, seed 42
2 — Rainbow portal dragon, seed 42
3 — Neon Buddha, seed 12345
4 — Psychedelic portrait, seed 42
5 — Alien road trip, seed 42
6 — Astronaut landscape, seed 42
7 — Rainbow queen, seed 314159

Scale 0.7 retains more of the original depth and pose while clearly applying the learned linework, palette, pattern density, and surreal graphic treatment. Scale 1.0 is more saturated and flatter; it may be preferable when maximum style strength matters more than preserving the base composition.

Plain-text prompt examples

Fibo also accepts ordinary prompt strings without VGL JSON expansion. These comparisons passed the text below directly to BriaFibo2Pipeline; no prompt-expansion step was used. Both columns use 1024 × 1024 resolution, 50 steps, guidance 5.0, an empty negative prompt, and the same prompt-specific seed. The stock column disables the adapter and omits the trigger. The LoRA column uses scale 0.7 and prepends fibo_psychpop_style,.

Prompt	Fibo stock — no trigger	LoRA 0.7 — with trigger
Cosmic white tiger, seed 4242
Alien desert camper, seed 314159

Prompts shown without the trigger used by the LoRA column:

a majestic white tiger walking through a moonlit jungle, its stripes glowing electric blue and hot pink, surrounded by giant tropical flowers, floating eyes, curling vines, stars, and concentric rainbow halos, psychedelic pop illustration with bold black outlines and luminous detail
a tiny retro-futurist camper van parked beside a crystal lake on an alien desert at sunset, two friendly travelers watching ringed planets rise above striped mountains, swirling neon clouds and strange flowering cacti, psychedelic poster art with saturated colors, intricate patterns, and bold outlines

The stock model already follows the explicit psychedelic wording. The LoRA makes the treatment substantially more saturated and graphic, adds denser patterning and neon contour work, and pushes both scenes closer to the training-set visual language. Structured VGL remains useful for precise control, but plain text is a viable inference mode for this adapter.

Volumetric subtype test

Alien road trip and astronaut landscape were also rerun at scale 0.7 with the structured-prompt subtype changed from digital illustration to smooth volumetric psychedelic 3D render. The scene description, objects, composition, palette, seed, resolution, steps, guidance, trigger, and LoRA scale were unchanged.

Prompt	Digital illustration / 0.7	Volumetric / 0.7
Alien road trip, seed 42
Astronaut landscape, seed 42

The volumetric wording produces smoother, glossier forms in the car scene and more dimensional lighting and terrain in the astronaut scene. The original subtype is more graphic and line-heavy, so both are retained as prompt-style options.

Diffusers usage

Access to the gated Fibo base model is required. Use a recent Diffusers checkout containing BriaFibo2Pipeline and BriaFibo2LoraLoaderMixin:

pip install -U git+https://github.com/huggingface/diffusers.git transformers accelerate peft safetensors
huggingface-cli login

import json
import torch
from diffusers import BriaFibo2Pipeline

base_model = "briaai/Fibo-2.0-dev"
lora_repo = "axiomofmind/Fibo-2.0-psychedelic-lora"

prompt = json.dumps(
    {
        "short_description": "A glowing mushroom with wavy blue and pink caps rises from a starry field against a black background.",
        "objects": [
            {
                "description": "a tall luminous mushroom with layered wavy blue and pink caps",
                "location": "center of the frame"
            }
        ],
        "background_setting": "A black nocturnal field fading into a star-filled void.",
        "lighting": {"description": "The mushroom casts blue and pink neon light."},
        "aesthetics": {
            "composition": "A centered vertical composition with ample negative space.",
            "color_palette": "electric blue, hot pink, violet, and black",
            "mood": "enchanted and otherworldly"
        },
        "style_medium": "fibo_psychpop_style, digital illustration",
        "context": "A surreal psychedelic night scene.",
        "artistic_style": "luminous contours, fluid shapes, high contrast",
        "text_render": []
    },
    separators=(",", ":")
)

negative_prompt = json.dumps(
    {
        "short_description": "",
        "objects": [{
            "description": "the primary subject",
            "texture": "flat, waxy, airbrushed, or lacking fine surface detail",
            "appearance_details": "blurred edges, malformed details, or simplified geometry"
        }],
        "background_setting": "featureless studio gradient with no material detail",
        "lighting": {},
        "aesthetics": {"artifacts": [
            "oversaturation", "color cast", "clipped highlights", "lens flare",
            "motion blur", "compression artifacts"
        ]},
        "photographic_characteristics": {
            "depth_of_field": "excessive bokeh obscuring the scene",
            "focus": "soft or out of focus"
        },
        "style_medium": "",
        "context": "",
        "artistic_style": ""
    },
    separators=(",", ":")
)

pipe = BriaFibo2Pipeline.from_pretrained(base_model, dtype=torch.bfloat16).to("cuda")
pipe.load_lora_weights(
    lora_repo,
    weight_name="pytorch_lora_weights.safetensors",
    adapter_name="psychpop"
)
pipe.set_adapters("psychpop", adapter_weights=0.7)

image = pipe(
    prompt=prompt,
    negative_prompt=negative_prompt,
    width=1024,
    height=1024,
    num_inference_steps=50,
    guidance_scale=5.0,
    generator=torch.Generator(device="cuda").manual_seed(42)
).images[0]
image.save("fibo_psychpop.png")

The included command-line example runs the same workflow with the built-in structured mushroom prompt:

python scripts/generate_fibo_lora.py \
  --scale 0.7 \
  --seed 42 \
  --output fibo_psychpop.png

Pass --prompt-json path/to/prompt.json to use another structured prompt. Its style_medium field must include fibo_psychpop_style.

Limitations

The adapter is intentionally strong and may flatten or simplify scenes at scales above 0.7.
The learned style is active whenever the adapter is enabled; the trigger nudges behavior but does not fully gate the style.
Fibo works best with structured VGL prompts. Plain text may be less predictable.
Test unfamiliar subjects and aspect ratios before production use.
Base-model limitations, biases, and licensing terms continue to apply.

Downloads last month: 153

Model tree for axiomofmind/Fibo-2.0-psychedelic-lora

Base model

briaai/Fibo-2.0-dev

Adapter

(3)

this model