<h2 id="an-analog-crt-and-retro-cyberpunk-aesthetic-lora">An Analog CRT &amp; Retro Cyberpunk Aesthetic LoRA</h2>This is a specialized style LoRA trained to inject authentic analog decay, retro CRT monitor physics, RGB chromatic aberration, and vintage broadcast interference into your generations.Unlike clean digital render models, this checkpoint transforms crisp digital assets into gritty, atmospheric photographs of physical CRT displays and analog media.Suggested Krea 2 Turbo Settings Model strength: 0.1-1.0 Samplers/schedulers:<ul><li>dpm_2_ancestral / ddim_uniform (or sgm_uniform)</li><li>exp_heun_2_x0 / ddim_uniform</li><li>euler / beta57 (or beta)</li></ul>Also note that you can push the model strength to 1.25 as well. I think beyond that gets a little too distorted, but try a variety of LoRA strength values! You can really dial things in to taste.<h2 id="what-it-does-best">📺 What It Does Best</h2><ul><li>Integrated Hardware Decay: Adds authentic horizontal scanline tearing, phosphor glow bleed, and rolling electromagnetic interference.</li><li>Color Fringing: Generates organic neon cyan and magenta RGB ghosting and chromatic aberration.</li><li>Text &amp; UI Fidelity: When dialed to the sweet spot (0.6–0.8 strength), it preserves complex typography and UI elements while wrapping them in raw retro atmosphere.</li></ul>While Krea 2 base model does understand the elements of this style, the effects it applies look a bit "pasted on top" in a fake way. This LoRA integrates those effects into the image. Krea 2 by itself also doesn't do a good job with the glow or haze from an old CRT screen, it's often missing. This LoRA corrects the overly digital, sterile, and bright look from the base model.<h3 id="advanced-tip:-negative-strength-(-1.0)">💡 Advanced Tip: Negative Strength (<code>-1.0</code>)</h3>Want a completely different look? Try setting the LoRA weight to a negative value like <code>-1.0</code>.Instead of adding analog CRT decay, inverting the weight acts as an "anti-CRT" filter that strips away the style bias. It forces the base model to pivot toward hyper-detailed, clean chrome/glass material textures and sharper structural geometry while keeping the chaotic glitch borders intact.

CRT Vaporwave

This version used a batch size of 1 and was trained to 3000 steps. It was a tough choice between this and another version that used an effective batch size of 16 trained to 2000 steps. This version was working a little better, especially when applied at higher strengths.

Files with SHA256: d08af66644a9df1deba5344b9a60450c4cc743b39f3af850c407736464608c7e