Max quality QWEN Edit 2509 outputs, minimal workflow, and lots of info

Multi-Image Notes<ul><li>Order: When inputting images, image one is on the right. So, add them right-to-left. They're labelled as well.</li><li>Use the right workflow: Because of the extra nodes, it's inconvenient 'bypassing' the 3rd or 2nd images correctly without messing it up. I'd recommend just using the three workflows separately rather than trying to do all three flexibly in one.</li><li>They take longer: A lot longer. The quality is maximal, but the 2-image one takes 3x longer than 1-image does, and the 3-image one takes 5x longer.<ul><li>This is because each new ref image gets added twice</li><li>Without using the reference latent nodes, they only take 2x and 3x longer because the images are only added once<ul><li>But, of course, the results are more blurry so it's a tradeoff</li><li>If you want to run fast tests, bypass the reference latent nodes to do test runs and then plug them back in when you want to do the high quality run</li></ul></li><li>Note that the first reference latent doesn't multiply time this way, so it's only the multi image workflows that are slow</li></ul></li><li>Spaghetti: The workflow connections look like spaghetti because each ref adds several nodes with cross-connections to other nodes. They're still simple, just not pretty anymore.</li><li>3-image reference order: In the 3-image version, the reference latents must be passed in reverse order into the second "combine conditioning" node. If you pass them the wrong way it will mess your output up completely. So, be careful if you're moving the workflow around.</li><li>Scaling: Reference images don't have as strict scaling needs. You can make them bigger or smaller. Bigger will make gens take longer, smaller will make gens faster.<ul><li>Make sure the main image is scaled normally, but if you're an advanced user you can scale the first image however you like and feed in a manual-size output latent to the k-sampler instead (as described further below in "Advanced Quality")</li></ul></li></ul>

Files with SHA256: 09683d80f7c24381b3dc588518a7648cd057bf843155850b0d2861c108392f0e