<h1 id="this-model-increases-&quot;quality&quot;-uk2zqr95e">This Model Increases "Quality"</h1>Quality here is defined by the model itself, as the training process I used involved reinforcing the model's "masterpiece" tag and various others across a large number of different example compositions.<h2 id="how-did-i-do-this-pw3pbcqms">How did I do this?</h2>I used the LECO training script on GitHub by P1atdev based on the LECO paper to train this model, the LECO training process generates an image at an arbitrary denoising strength, and then trains on the difference between the output of the model when prompted for a concept, and when unprompted. this allows the model to alias tags, words, concepts, or phrases to any arbitrary prompt, in this case, I aliased<pre><code>masterpiece, best quality, newest, absurdres, highres, high quality, highly detailed</code></pre>the "quality paragraph" to<pre><code></code></pre>so I effectively was training it to always produce images that looked as though it had been prompted for the "quality paragraph"<h2 id="donations-i74w95u9f">Donations</h2>Speaking of training, training models is expensive and I run training on my own private server, if you like what I do, consider supporting the development here!<a target="_blank" rel="ugc" href="https://ko-fi.com/yolup">https://ko-fi.com/yolup</a><h2 id="key-benefits-m0xt8qiow">Key Benefits</h2><ol><li>One of my favorite key benefits of this approach is that it makes "quality" a modular and controllable thing. adding weight to the quality tag has a somewhat enigmatic effect on the output however, this LoRA/LECO has very clearly defined and comprehensible changes that you can control the severity of by altering the weight of the LoRA/LECO which is an intended operation (as opposed to weighted prompts being a hack applied to the attention layer that doesn't always have the desired effect)</li><li>The other benefit is that this LoRA/LECO doesn't use up the tokens that the "quality paragraph" uses! eating up the context window forces the backend you are using, whether it be A1111, InvokeAi, or ComfyUI to add invisible BREAKs in attention that damage the overall coherence of the prompt you are constructing and can lead to other unintended consequences.</li></ol><h2 id="quirks-pshqn8st0">Quirks</h2>I have tested it in many instances and the only things I have routinely noticed are the following 1) if unspecified, it tends to make figures feminine 2) It seems to move towards a "sharp" look at higher weights3) it greatly affects the composition, making it not ideal if you wish to simply improve an earlier generation4) I've noticed that it seems to increase the exposure of an image5) it appears to make the model more reliable in some way, turning the CFG scale below and above the base model operating range is something I have done and it does appear to recover normal operation, which I find strange.

NoobAI-XL Quality Slider

Trained on NoobAIXL Epsilon 1.1 rank 4

NAI_QualityV1.1.safetensors

Epsilon v1.1

This Model Increases "Quality"Quality here is defined by the model itself, as the training process I used involved reinforcing the model's "masterpiece" tag and various others across a large number of different example compositions.How did I do this?I used the LECO training script on GitHub by P1atdev based on the LECO paper to train this model, the LECO training process generates an image at an arbitrary denoising strength, and then trains on the difference between the output of the model when prompted for a concept, and when unprompted. this allows the model to alias tags, words, concepts, or phrases to any arbitrary prompt, in this case, I aliased: masterpiece, best quality, newest, absurdres, highres, high quality, highly detailedso I effectively was training it to always produce images that looked as though it had been prompted for the "quality paragraph"DonationsSpeaking of training, training models is expensive and I run training on my own private server, if you like what I do, consider supporting the development here!<a target="_blank" rel="ugc noopener noreferrer nofollow" href="https://ko-fi.com/yolup">https://ko-fi.com/yolup</a>Key Benefits<ol><li>One of my favorite key benefits of this approach is that it makes "quality" a modular and controllable thing. adding weight to the quality tag has a somewhat enigmatic effect on the output however, this LoRA/LECO has very clearly defined and comprehensible changes that you can control the severity of by altering the weight of the LoRA/LECO which is an intended operation (as opposed to weighted prompts being a hack applied to the attention layer that doesn't always have the desired effect)</li><li>The other benefit is that this LoRA/LECO doesn't use up the tokens that the "quality paragraph" uses! eating up the context window forces the backend you are using, whether it be A1111, InvokeAi, or ComfyUI to add invisible BREAKs in attention that damage the overall coherence of the prompt you are constructing and can lead to other unintended consequences.</li></ol>QuirksI have tested it in many instances and the only things I have routinely noticed are the following 1) if unspecified, it tends to make figures feminine 2) It seems to move towards a "sharp" look at higher weights3) it greatly affects the composition, making it not ideal if you wish to simply improve an earlier generation4) I've noticed that it seems to increase the exposure of an image5) it appears to make the model more reliable in some way, turning the CFG scale below and above the base model operating range is something I have done and it does appear to recover normal operation, which I find strange.Hide

Files with SHA256: 30e8248703974e581331dcdddc62b818d8559846a2d4a21ec624f6450b09bb71