Loading
Vous avez un workflow ComfyUI ? Obtenez une documentation lisible directement dans votre navigateur.Rendre un workflow lisible.Essayer maintenant ↗
Typnosis is a typography LoRA trained on a varied set of typographic images, including clean lettering, 3D text, material based type, embedded text, and more experimental compositions. The training covers different...
Source du modèle
Description de la source
Typnosis is a typography LoRA trained on a varied set of typographic images, including clean lettering, 3D text, material based type, embedded text, and more experimental compositions. The training covers different approaches to typography rather than focusing on a single visual style.
Example below
Usage
At lower strengths, Typnosis mostly follows the prompt and adds a lighter typographic influence. As the strength increases, it starts affecting the whole image more strongly. At the highest strengths, it can become much less prompt faithful and much more stylized.
No trigger word is required.
There is no single recommended strength. The useful range depends on what you want from the LoRA.
Sources
1 sourceArtefacts du modèle
1 artefactTypnosis_Krea2.safetensors
safetensors · BF16 · 436 MB · SHA-256 818605ee6c3f…5ae2 · Civitai
TéléchargerExtraits de sources
1 extraitSource context: 321 downloads · Type LORA · Base model Krea 2
If prompt fidelity matters more, keep the strength lower. If you care more about the typography itself, pushing the strength higher can produce more stylized results.
Detailed prompts still matter. You can describe the text, layout, material, depth, environment, and how the typography should interact with the scene.
STR1 to STR2 is the most balanced range. Higher strengths push the style further and may distort the text. At STR4 and above, compositions tend to become simpler and different prompts start converging toward a more similar look.
Showcase
The showcase is not cherry picked.
I used the first generations from my tests with the same seed and intentionally kept imperfect results as well. Each prompt was generated from STR0 to STR5 instead of selecting the best strength for each image.
This should give a more realistic idea of what the LoRA does, how it changes across strengths, and whether it fits what you are looking for before you test it yourself.
