🎴 Merging Lora, Textual Inversion and Fine-tuned models
Now is it time to put everything we have discussed into one image. We will use the same prompt and parameters as in our previous lecture of negative textual inversion and Lora. This will help to emphases the power of the a few tweaks can make. We will be using this model here Realistic Vision v1.3
Prompt: portrait photograph of Madison Beer as Pocahontas, young beautiful native american woman, perfect symmetrical face, indigenes feather jewelry, traditional handmade dress, armed female hunter warrior, (((wild west))) environment, Utah landscape, ultra realistic, concept art, elegant, ((intricate)), ((highly detailed)), depth of field, ((professionally color graded)), soft ambient lighting, dusk, 8k, art by artgerm and greg rutkowski and alphonse mucha <lora:epiNoiseoffset_v2:0.5> <lora:iu_V35:0.3> <lora:analogDiffusionLora_v1:0.2>
Negative prompt: easynegative ng_deepnegative_v1_75t badhandv4 bad-picture-chill-75v
Steps: 100 Sampler: Euler a CFG scale: 7 Seed: 6248814 Size: 512x768 Model hash: c35782bad8 Model: Realistic_realisticVisionV20_v13
Hashes: {"embed:bad-picture-chill-75v": "7d9cc5f549", "embed:badhandv4": "5e40d722fc", "embed:ng_deepnegative_v1_75t": "54e7e4826d", "lora:epiNoiseoffset_v2": "81680c064e", "lora:iu_V35": "c9598ba347", "lora:analogDiffusionLora_v1": "b3db478ff8"}
Now let's compare the results of all our images to see how we have transformed our simple prompt.
Again, using all the same prompt and parameters, from top left to bottom right we have:
- Model SDv1.5 with no negative prompts
- Model SDv1.5 with negative embedding
- Model SDv1.5 with negative embedding and LoRAs
- Custom Model with negative embedding and LoRAs