💡 Search "huggy" to find all my models on site.💡 Models are updated frequently, so be sure to check back and use the latest version.💡 Civitai switched its backend from A1111 to ComfyUI earlier this year (2024). That means you can't reuse prompts or seeds from last year to generate the same image. It's best to remove all weight values and start fresh.The latest version was trained on 1280 × 960 (4:3 ratio) but when generate on Civitai with 1216 × 832 (7:11 ratio), the mismatch can cause distortions like long necks, especially on Pony XL and illustrious base model. Another reason to move to flux.I hate to say it, but most my loras won't work with DucHaiiten Pony base model.This is an early experimental project which is also one of my first 3 loras made on June 2024, trained with 53 images with ratio of 1024 x 1024, featuring realistic drawing, minimal flat animation and sharpie illustration. After three months of iterations, it developed into various versions with additional realistic painting and photo realistic style with around 300 images. After 29 lora of experimentations I come back to this original lora and think about starting all over again.📒 Useful Tips:<ul><li>Make sure you are using correct base model, a pony lora won't work consistently on SDXL / SDXL illustrative</li><li>All Civitai showcase images are generated with A1111 backend. Remove all tags weight values in prompt if you are using ComfyUI, ignore other settings like CFG, steps.</li></ul><ul><li>As the lora name tells, the source images are selected for pastel and flat styles. To achieve this you need additional tags like "anime, pastels drawing, flat drawing" to emphasise the effect (this does not apply to Version 2). But I usually use it as a character lora as they are looking good.</li></ul><ul><li>There's no universal prompt / lora weight for every generation. If you're using anime style / character lora with realistic checkpoint, starting with a lower lora weight like 0.7. Ignore the lora weight in sample images, due to a Civitai bug, image generated online will display lora weight of 1 regardless the actual value.</li></ul><ul><li>For close-up portrait, start steps value of 22; for full body shot or couple shot, start steps value of 34.</li></ul><ul><li>The training files are mostly focus on solo character, which means if you intent to generate a image with couple, you might get unexpected result. To fix this you need to use an additional lora trained with multiple people (like ExpressiveH) or a pose lora. A good multiple people lora should contain as many poses as it can with couple, threesome or a group of people and it is definitely a tough job.</li><li>This lora will rely on negative prompts to fix hands, eyes, and accessories, check my showcase images for reference.</li></ul>🖍️ Changes in Version V2:1) Removed source images of sharpie illustration style to improve eye drawing quality.2) Increased training source resolution.3) Reduced training steps down to 1000.📒 MirroringI’ve noticed a few sites, like Modelslab and yodayo, using bots to mirror my models. SeaArt in particular shows big numbers for downloads and generations.<ol><li>I don’t mind if you mirror my models — sites like civitaiarchive have always done that.</li><li>I do mind if you block users from downloading the models and force them onto a paid online service.</li><li>I do mind if you claim copyright or give users ownership of the generations. These are open-source models, trained on open-source bases with limited-ownership data. No one — not even me — should claim copyright. Credit is all that’s needed.</li></ol>

Pastelflat 🖍️ by Huggy

Image 53networkDim:32networkAlpha:32

Pastelflat_Pony_V02-000005.safetensors

Pastelflat 🖍️ from Huggy

V01E05

V02E10

V02E06

V02E04

V01E20

V01E17

V01E14

V01E11

V01E07

pastelflat V00E10

Files with SHA256: cdb1783f1f23b647c1c3b1b6339d9369b631c366aeea7b666889ce6b7d3a78ea