📱 GalaxyAce LoRA
🎬 MiniMax H3 added
Model: MiniMax H3
Recommended LoRA Strength: 1.00
Trigger Word: Not required
📷 Krea 2 added
Model: Krea 2 (trained on Krea-2-Raw, works on Raw & Turbo)
Recommended LoRA Strength: 1.0 for in-dataset scenes — up to 2.0 (see Strength Behavior)
🎬 LTX-2.3 added
Model: LTX-2.3
Recommended LoRA Strength: 1.00
Trigger Word: Not required
In LTX 2.3 T2V&I2V ComfyUI Workflow, LoRA is connected immediately after the checkpoint node inside the subgraph
🎬 LTX-2 added
Model: LTX-2
Recommended LoRA Strength: 1.00
Trigger Word: early 2010s low-end android phone footage
This LoRA replicates the early 2010s video style of a Samsung Galaxy Ace smartphone.
The training videos include streets, parks, everyday scenes, casual portraits, and real-life moments typical for low-end Android devices.
The LoRA transfers:
color reproduction of the Galaxy Ace camera
sensor noise & compression artifacts
dynamic range and exposure behavior typical for small sensors
lens softness and subtle edge blur
subtle motion jitter and handheld feel
overall “early 2010s low-end Android phone video” aesthetics
⭐ How to Use
LoRA Strength: 1.00
Base Model: LTX-2
Best with natural scenes: street videos, parks, cafes, casual portraits, travel clips, everyday life
Adds subtle imperfections and low-fi realism for video
Recommended workflow: official ComfyUI LTX-2 Text-to-Video pipeline
🧪 Notes
Ideal for: street videos, diary-style clips, lifestyle captures, travel footage, nature
Preserves realism, avoids stylization, reproduces small-sensor behavior
Enhances authenticity while keeping motion, lighting, and colors faithful to early 2010s low-end Android look
📷 Z-Image-Turbo (Photos)
Model: Z-Image-Turbo
Recommended LoRA Strength: 1.20
⭐ How to Use
LoRA Strength: 1.20
Base Model: Z-Image-Turbo
Works perfectly with natural, simple prompts
Adds subtle imperfections and real-photo feel of a low-megapixel phone camera
🧪 Notes
Best results: street photos, portraits, pets, travel shots, nature, everyday objects
Maintains realism, avoids stylization
Recreates small-sensor characteristics of the original device
Description
📷 Krea 2
Model: Krea 2 (trained on Krea-2-Raw, works on Raw & Turbo)
Recommended LoRA Strength: 1.0 for in-dataset scenes — up to 2.0 (see Strength Behavior)
Trigger Word: Not required
This LoRA brings the Galaxy Ace / cursed early-digital look to Krea 2 — harsh flash, washed colors, lo-fi grain, panelka & village energy.
It also doubles as a Realism Detailer: at lower strengths it strips Krea's "too-perfect" look and makes skin and lighting believable.
⭐ How to Use
LoRA Strength:
– 1.0 for natural everyday scenes that match the training vocabulary
– 1.25–1.45 as a Realism Detailer (subtle; keeps composition, removes the plastic look)
– 1.50–2.0 for full Galaxy Ace cursed style, or for unfamiliar / exotic prompts (see below)
Base Model: Krea 2 — Turbo (~8 steps) or Raw (~30 steps), FlowMatch, guidance ~4 Best with natural, simple prompts. Add harsh flash for night, motion blur for action.
🧪 Strength Behavior (important)
The style binds strongest to concepts that were in the training set:
— Prompts close to the dataset (everyday / casual scenes) — the look applies fully at ~1.0
– Prompts far from the dataset (exotic, fantasy, unusual subjects) — crank to 1.25–2.0 to push the style onto unfamiliar content
This is normal for a style LoRA trained with descriptive captions — not a defect. Use it as a dial: low for realism, high for full cursed or out-of-distribution scenes.
🧪 Notes
– Trained with Ostris AI-Toolkit, LoRA rank 16, ~1250 steps on RTX Pro 6000 (96GB) — under 30 minutes of training
– Trigger-less, runs hot — don't be afraid to go above 1.0
– Two tools in one: Realism Detailer (low strength) + full Galaxy Ace style (high)
FAQ
Comments (11)
OMFG. Congrats on this, this LoRA is absolutely insane. Zero stock photo or fake vibe. I tested it from 1.40 up to 1.80 strength and the results are incredibly convincing, This is unreal!!!!
Would it be possible for you to make a LoRA that simulates photos taken between 2019 and 2021? Like the iPhone 11 to 13 era, or Samsung? Same vibe as that GalaxyAce you did, but obviously a bit more "modern"?
Man, congrats. You literally broke Krea 2 in what, less than 2 days since the model dropped???
The effect isnt much prominent for KREA 2 in strength 1-1.6 but then beyond that its great, combined with Realism V!, its a peak thing. Great work man 👌👌
do u mean realism version 1? what lora strength u use for each
I'd love to see your workflow for the LTX 2.3 video previews. Any chance you can upload it? There's no blur at all in movement and I'm curious how you managed it and what I'm missing in my own workflows.
This is the basic workflow comfyui for ltx 2.3
Of all the realism loras ive tested on civit.red for Krea2, this one is by far the best and most realistic. It truly pushes Krea2 away from that harsh synthetic look that it clings onto incredibly hard. I'm using a weight of 1.55 and love it. Using trigger words like candid and amateurish helps a ton.
any chance for an ideogram version?
Possible
Hey, I think your krea2 lora causes most text characters to become cyrillic. Its actually crazy, must be the tagging vs visual correlation. I've never seen any other model learn that good :)
When i use in Krea 2,all images look like night time flash photography, why? even i mention day time, natural soft light in prompt, it produces night look only, any keywords to use this lora?
The LoRa is great and I'm loving the realism, camera noise and amateur feel but sadly it keeps creating weird intrincate and off putting "dirty patterns" specially flowers or similar shapes on surfaces such as walls, ceilings, curtains... Specially on the background. Something in the dataset I assume.
This effect is even worse if you stack the LoRa with other LoRas
Details
Files
galaxyace_krea2.safetensors
Mirrors
galaxyace_krea2.safetensors
galaxyace_krea2.safetensors
galaxy_ace.safetensors
galaxyace_krea2.safetensors
galaxyace_krea2.safetensors
galaxyace_krea2.safetensors
galaxyace_krea2.safetensors
galaxyace_krea2.safetensors
galaxyace_krea2.safetensors
galaxyace_krea2.safetensors
GalaxyAce_KREA2_[aiguild].safetensors