issue205:stable_diffusion
Différences
Ci-dessous, les différences entre deux révisions de la page.
Les deux révisions précédentesRévision précédenteProchaine révision | Révision précédente | ||
issue205:stable_diffusion [2024/06/14 07:19] – d52fr | issue205:stable_diffusion [2024/06/14 09:24] (Version actuelle) – auntiee | ||
---|---|---|---|
Ligne 2: | Ligne 2: | ||
Why is this very useful? – You can copy the entire workflow and paste it on the same page allowing you to make changes without starting on a new workflow. You don’t lose anything in the original version, and can switch back and forth as needed to run only the active one. ** | Why is this very useful? – You can copy the entire workflow and paste it on the same page allowing you to make changes without starting on a new workflow. You don’t lose anything in the original version, and can switch back and forth as needed to run only the active one. ** | ||
+ | |||
+ | Cette fois-ci, nous continuons avec ComfyUI. Nous commencerons par une fonction utile, puis nous créerons un flux de travail parallèle et en série. L' | ||
+ | |||
+ | Pourquoi est-ce très utile ? Vous pouvez copier l' | ||
+ | |||
**Next we will create parallel processes in a workspace. Consider that you may want to create an image with each of the four seasons. Essentially you need to make four identical workflows all starting from the same beginning. Start with the original image workflow, copy the Ksampler and paste it with Ctrl-Shift and V three times, each with the same starting connections. All that’s needed then is to add a CLIP Text Encode (Prompt) for each season with the prompt word summer, fall, winter or spring. Combining each season with the original positive prompt via a new Conditioning (Combine) Node and using that as the positive Prompt to the KSampler allows you to get the desired output images. | **Next we will create parallel processes in a workspace. Consider that you may want to create an image with each of the four seasons. Essentially you need to make four identical workflows all starting from the same beginning. Start with the original image workflow, copy the Ksampler and paste it with Ctrl-Shift and V three times, each with the same starting connections. All that’s needed then is to add a CLIP Text Encode (Prompt) for each season with the prompt word summer, fall, winter or spring. Combining each season with the original positive prompt via a new Conditioning (Combine) Node and using that as the positive Prompt to the KSampler allows you to get the desired output images. | ||
While the workflow looks confusing, it’s four copies of the same thing except for the added Prompt via the Combining Node. We can do something similar but as a series. We previously up-scaled an output, only to find it made the image bigger but with no higher resolution. ** | While the workflow looks confusing, it’s four copies of the same thing except for the added Prompt via the Combining Node. We can do something similar but as a series. We previously up-scaled an output, only to find it made the image bigger but with no higher resolution. ** | ||
+ | |||
+ | Nous allons ensuite créer des processus parallèles dans un espace de travail. Imaginez que vous souhaitiez créer une image pour chacune des quatre saisons. En fait, vous devez créer quatre flux de travail identiques, qui partent tous du même point. Commencez par le flux de travail de l' | ||
+ | |||
+ | Bien que le flux de travail semble déroutant, il s'agit de quatre copies de la même chose, à l' | ||
+ | |||
**To solve that issue, we will chain three KSampler nodes with an Upscale Latent Node taking the output Latent from the first KSampler node to the second KSampler node via a Latent Upscale By node; we will repeat the same steps from the second KSampler node to the third KSampler node. I used a scale_by 2.0 for both Latent Scale By settings. In each of the above examples, if the first image doesn’t look good you can simply stop the workflow via the Queue Prompt manager before it finishes to save time which is particularly important if it’s a slow process. | **To solve that issue, we will chain three KSampler nodes with an Upscale Latent Node taking the output Latent from the first KSampler node to the second KSampler node via a Latent Upscale By node; we will repeat the same steps from the second KSampler node to the third KSampler node. I used a scale_by 2.0 for both Latent Scale By settings. In each of the above examples, if the first image doesn’t look good you can simply stop the workflow via the Queue Prompt manager before it finishes to save time which is particularly important if it’s a slow process. | ||
Ligne 12: | Ligne 22: | ||
Next time, we will look at yet another interface UI version named Fooocus. It doesn’t have the flexibility of ComfyUI to control each step but for simplicity it’s hard to beat. ** | Next time, we will look at yet another interface UI version named Fooocus. It doesn’t have the flexibility of ComfyUI to control each step but for simplicity it’s hard to beat. ** | ||
+ | |||
+ | Pour résoudre ce problème, nous enchaînerons trois nœuds KSampler avec un nœud Latent Upscale prenant la sortie Latent du premier nœud KSampler vers le deuxième nœud KSampler via un nœud Latent Upscale By ; nous répéterons les mêmes étapes du deuxième nœud KSampler vers le troisième nœud KSampler. J'ai utilisé un scale_by de 2.0 pour les deux paramètres Latent Scale By. Dans chacun des exemples ci-dessus, si la première image n'est pas bonne, vous pouvez simplement arrêter le flux de travail via le gestionnaire Queue Prompt avant qu'il ne se termine pour gagner du temps, ce qui est particulièrement important si le processus est lent. | ||
+ | |||
+ | Le résultat montre le flux de travail avec apparemment trois images exactes. Cependant, j'ai recadré les têtes des trois sorties et vous y voyez une différence notable. Les images sont non seulement plus grandes, mais leur résolution est plus élevée, car davantage d' | ||
+ | |||
+ | La prochaine fois, nous examinerons une autre version de l' | ||
issue205/stable_diffusion.1718342396.txt.gz · Dernière modification : 2024/06/14 07:19 de d52fr