Ideogram 4.0 LoRA Collection

📧 Free 5-page LoRA training guide → https://batmanosama.kit.com/9fee60068c

A curated collection of style LoRAs trained on Ideogram 4.0. This repository will be regularly updated with new aesthetic styles and specialized LoRAs.

To be used with the Ideogram 4.0 model. As this model is quite new, things are still experimental.

If you're looking for a ComfyUI workflow for Ideogram with lora support checkout this Reddit post

Available LoRAs

VHS_RALLY_95 v1 (1995 Hi8 camcorder found-footage aesthetic)

Download: vhs_rally_95.safetensors in the Files & versions tab

A style LoRA that transforms any scene into authentic 1995-era Hi8 camcorder found footage. Think grainy amateur documentary footage recorded on a consumer Handycam — complete with VHS artifacts, timestamp overlays, flat underexposed lighting, and that unmistakable low-budget realness.

Training details:

Base model: Ideogram 4.0 (FP8)
Trainer: ai-toolkit (flowmatch)
Dataset: 65 images (Mario Kart characters in 1995 found-footage documentary style)
Rank: 32 | Alpha: 32
Learning rate: 1.5e-5
Steps: 2,500
Format: Native ai-toolkit safetensors (directly compatible with ComfyUI)

Recommended settings:

LoRA strength: 0.8–0.9 (model and clip)
Sampler: Euler
Steps: 20
CFG: 7.0 base, with CFG Override to 3.0 at 70–100%
Resolution: 1024×768 (landscape)

Trigger/approach: This is a style-only LoRA — no trigger word needed. Simply load the LoRA and describe your scene. The style will apply the 1995 camcorder aesthetic to whatever you generate. Works best with prompts that describe documentary-style, candid, or action scenes.

Sample prompts:

Authentic 1995 low-budget amateur video still from underground go-kart racing documentary. Mario costume driving a battered red-and-white go-kart on a muddy dirt track. Real mud spray, real costumes, cheap props. Overcast sky, flat natural lighting. VHS timestamp in corner. Low resolution, found footage, amateur documentary.

Close-up shot of a rusted go-kart engine covered in mud and grease, handheld camcorder footage from 1995. Shallow depth of field, auto-focus hunting. VHS tracking lines, timestamp overlay. Low resolution, found footage.

Wide shot of a rainy pit stop area, makeshift garage with tools scattered on concrete floor, 1995 amateur documentary footage. Flat lighting, VHS artifacts, chromatic aberration. Low resolution, found footage.

Showcase:

Ektachrome Style v1 (vintage 1960s/70s Kodak Ektachrome film aesthetic)

Download: ektachrome_style_v1.safetensors in the Files & versions tab

Add the trigger word ektachrome in the art_style section of the style description, to be used in combination with appropriate aesthetics and mediums. All samples are generated with the following style_description:

{
  ...
  "style_description": {
    "aesthetics": "Desolate, industrial, stark",
    "lighting": "Bright, direct overhead sunlight",
    "medium": "Photograph",
    "photo": "Photograph. Film grain visible. Wide-angle.",
    "art_style": "ektachrome"
  }
  ...
}

Other examples:

"aesthetics":"Desolate, industrial, stark"
"aesthetics":"Moody, cinematic, atmospheric"
"aesthetics":"Stylized, colorful, cinematic"
"aesthetics":"Cold, scientific, archival"
"aesthetics":"Vintage, 1960s, Kodachrome-like"

"art_style":"ektachrome, vintage 1960s Kodak Ektachrome film emulation"
"art_style":"ektachrome, industrial documentation style with heavy film grain"
"art_style":"ektachrome, archival scientific photography with faded colors"
"art_style":"ektachrome, military test site documentation aesthetic"

BLOCK_MISSION_2004 v1 (2004 PS2 real-time framebuffer aesthetic)

Download: block_mission_2004.safetensors in the Files & versions tab

A style LoRA that transforms any scene into an authentic 2004-era PS2 game screenshot. Low-poly 3D geometry, compressed textures with visible DCT blocks, vertex lighting with hard banding edges, aggressive jaggies from 480i interlaced output, and the unmistakable warm orange color grading of West Coast crime games.

Training details:

Base model: Ideogram 4.0 (FP8)
Trainer: ai-toolkit (flowmatch)
Dataset: 66 images (17 prompts x 4 images from Google Flow Nano Banana Pro)
Rank: 32 | Alpha: 32
Learning rate: 1e-4
Steps: 2,500
Optimizer: AdamW 8-bit
Format: Native ai-toolkit safetensors (directly compatible with ComfyUI)

Recommended settings:

LoRA strength: 0.8–0.9 (model and clip)
Sampler: Euler
Steps: 20
CFG: 7.0 base, with CFG Override to 3.0 at 70–100%
Resolution: 1024×768 (landscape)

Trigger/approach: This is a style-only LoRA — no trigger word needed. Simply load the LoRA and describe your scene. The style will apply the 2004 PS2 framebuffer aesthetic to whatever you generate. Works best with prompts that include these visual descriptors:

2004 PS2 real-time cutscene, low-poly, low-res compressed textures,
vertex lighting with visible banding, baked bloom, jaggies, no anti-aliasing.
PS2 framebuffer 4:2:2 color. Interlaced 480i combing, MPEG-2 DCT block artifacts.
Warm orange color grading. No modern photorealism, no remake polish.

Sample prompts:

2004 PS2 real-time cutscene, low-poly West Coast crime game. Protagonist on a rooftop overlooking a sun-drenched city skyline with palm trees, a purple lowrider parked nearby on the concrete. Orange smog haze, vertex lighting banding on building faces, visible jaggies on all edges. Interlaced 480i scanlines, MPEG-2 compression artifacts. Warm color grading. Low-res compressed textures, no anti-aliasing.

2004 PS2 real-time cutscene, low-poly. Interior of a fast food restaurant with orange and yellow walls, protagonist at the counter. Fluorescent lighting with vertex banding on floor tiles. Low-poly character models, compressed menu board textures. Jaggies on counter edges. Interlaced 480i, MPEG-2 artifacts. Warm orange tone.

2004 PS2 real-time cutscene, low-poly. Nighttime police chase downtown, protagonist fleeing on foot between buildings. Police cruisers with red and blue lights, helicopter spotlight from above. Neon signs on storefronts. Heavy jaggies, vertex lighting on wet pavement, interlacing scanlines. Dark blues and purples with neon highlights.

Showcase:

Disclaimer and Important notes

The LoRAs in this repository are created for research purposes only and are not intended for commercial use. Some are based on TV series which are protected by copyright. Users utilize these models at their own risk. Users are obligated to comply with copyright laws and applicable regulations. The models have been developed for non-commercial purposes, and it is not my intention to infringe on any copyright. I assume no responsibility for any damages or legal consequences arising from the use of these models. no more no less

Downloads last month: 378

Model tree for jmanhype/Ideogram-4.0-LoRAs

Base model

ideogram-ai/ideogram-4-fp8

Adapter

(20)

this model