Scroll near the bottom for comments on the merged version (my original model merged with three others) <h3 id="photo-focused-model-with-a-narrow-aesthetic-but-versatile-subject-matter.-efpkp6og7">Photo-focused model with a narrow aesthetic but versatile subject matter.</h3>Mostly SFW model - only capable of the tamest NSFW kind of imagery.The starting point is my LoRA <a target="_blank" rel="ugc" href="https://civarchive.com/models/490059/eldritch-candid-photography">Eldritch Candid Photography</a> which I wanted to push a little further and so merged into a checkpoint with alternate versions and various other LoRAs to get this film aesthetic: slightly compressed dynamic range, generalized finer details, and heavy grain (as of old ISO 800 film) but with an overall sense of authenticity and convincing real-world presence. The benefits of the model, over the original LoRA, are greater subject matter flexibility and prompt coherence.I always prompt for 'photograph' as in 'cinematic photography' or 'product photograph' etc. Without which the model may lean toward a kind of painterly art output. Which can be quite nice, but its real raison d'être, for me, is photographic.<h3 id="upscale-or-refine-your-output:-z845q2ox0">Upscale or refine your output:</h3>Maybe the best results you can get from this model include upscaling the image, again using the same Eldritch Photography model. At larger resolutions, the grain/noise looks much more natural, doesn't overwhelm the imagery so much, while retaining the same lighting, atmosphere, and analog realism. <a target="_blank" rel="ugc" href="https://civarchive.com/posts/4073332">Here's an upscaled unicorn for example</a>.Alternatively, you can use another model in upscaling, or as a refiner. My preference is the first approach, upscaling with this same model. But in both approaches, I highly value the overall tonality and aesthetic that Eldritch Photography brings.Important note about LoRAs: the model is a crapshoot in how well LoRAs will work with it. Some work great, but may need lower strengths than you'd use on the base SDXL, others work fine at 1.0 and yet others are kinda garbage. It's a little hard to predict, but feel free to see what you get. To my taste, more general-purpose fine-tune checkpoints are better suited for loading up style LoRAs. I just use this model for its distinctive photo style.<hr /><h3 id="about-the-new-eldritch-photography-merge-version-ivdihecht">About the new Eldritch Photography Merge version</h3>The original Eldritch Photography model is noisy - maybe too much for some. Personally, I'm very fond of it. But it's also constrained. I don't have the computing resources (nor want to invest in them at this time) to train the many thousands of images that would be required to achieve a more versatile, general-purpose checkpoint. And I wanted to see what it might look like if I were to put in such time. Hence, this merge:Equal parts of <a target="_blank" rel="ugc" href="https://civarchive.com/models/133005?modelVersionId=456194">Juggernaut</a>, <a target="_blank" rel="ugc" href="https://civarchive.com/models/267242/proteus?modelVersionId=666604">Proteus</a>, and <a rel="ugc" href="https://civarchive.com/models/261804/xenodiffusion?modelVersionId=668785">XenoDiffusion</a> and a strong influence of Eldritch Photography, gets us this more versatile model. Output remains grainy, like low-light / high ISO film, and the dynamic range is reduced, flattening highlights and shadows somewhat, for an overall organic feel, just not quite so strongly as my original model. The other three original fine-tuned checkpoints also bring back more versatility, allowing this merge to explore more art aesthetics and work better with style LoRAs.

Eldritch Photography

eldritchPhotography_v10.safetensors

v1.0

Eldritch_Merge

Photo-focused model with a narrow aesthetic but versatile subject matter.Sample image generation data available on<a target="_blank" rel="ugc noopener noreferrer nofollow" href="https://civitai.com/models/549213?modelVersionId=611042"> the model's original post at the other site</a>. I highly recommend use of the DPM++ samplers. The starting point is my LoRA <a target="_blank" rel="ugc noopener noreferrer nofollow" href="https://civitai.com/models/490059/eldritch-candid-photography">Eldritch Candid Photography</a> which I wanted to push a little further and so merged into a checkpoint with alternate versions and various other LoRAs to get this film aesthetic: slightly compressed dynamic range, generalized finer details, and heavy grain (as of old ISO 800 film) but with an overall sense of authenticity and convincing real-world presence. The benefits of the model, over the original LoRA, are greater subject matter flexibility and prompt coherence.I always prompt for 'photograph' as in 'cinematic photography' or 'product photograph' etc. Without which the model may lean toward a kind of painterly art output. Which can be quite nice, but its real raison d'être, for me, is photographic.Upscale or refine your output:Maybe the best results you can get from this model include upscaling the image, again using the same Eldritch Photography model. At larger resolutions, the grain/noise looks much more natural, doesn't overwhelm the imagery so much, while retaining the same lighting, atmosphere, and analog realism. <a target="_blank" rel="ugc noopener noreferrer nofollow" href="https://civitai.com/posts/4073332">Here's an upscaled unicorn for example</a>.Alternatively, you can use another model in upscaling, or as a refiner. My preference is the first approach, upscaling with this same model. But in both approaches, I highly value the overall tonality and aesthetic that Eldritch Photography brings.Important note about LoRAs: the model is a crapshoot in how well LoRAs will work with it. Some work great, but may need lower strengths than you'd use on the base SDXL, others work fine at 1.0 and yet others are kinda garbage. It's a little hard to predict, but feel free to see what you get. To my taste, more general-purpose fine-tune checkpoints are better suited for loading up style LoRAs. I just use this model for its distinctive photo style.

Files with SHA256: 45a25b436bed08cf64bd26c097f156b7a3fd04a011417831e805a5ab4863425e