A workflow for ideogram using a mixed method optimized for speed with maintaining as much quality as possible - achieving in my opinion some of the best quality by far. I have tested it with various high resolution (8Mpx to 33Mpx, 4K to 8K, various extra wide aspect ratios as well). Included are some results made with this method. The only downside I've noticed is text with no background getting sub-standard when compared to the normal Ideogram workflow - for that I've included an alternate version. Instructions: Set the models, LoRAs you use (I use the standard fp8_scaled models, if yours is different, adjust as needed). Set width, height, upscale multiplier. Use the prompt builder (or the Manual JSON) to build the prompt. The included alternative workflow is fabricated so it keeps the superior fine text rendering abilities intact. To use it, set "Use method2" to true. You can batch both method 1 and method 2 for comparison. Method 2 is about 20% slower (though nearing 15Mpx it's getting close to the same speed).

Files with SHA256: 1cf24974c4fc04969e2d27f6dd5571c15773aaa8c6f0183adedd30def4d9b502