Flying Dog

Your browser does not support HTML5 video. BUY NOW AI Superpower for Photos​hop Your Editing Workflow will never be the same again Next level AI editor: flying dog AI Studio - Video Four powerful AI connectors: Three for Stable Diffusion and DALL-E 2 Use your own Stable Diffusion Server Text2Image, Inpainting, Outpainting and Image2Image Modifiers Library Working on Selection Finetuned Dreambooth models Background Tasks, Multi Server Management and more BUY now This software speeds up the work of designers like a turbocharger. DER SPIEGEL, 46, page 72 Works FantasicDefinitely a great tool for Outpainting. Tom Traubitz Fantasic AppThis is by far the best solution out there for Stable Diffusion at this time. Christopher Marin +++ Good news: Photoshop 24.3 solves compatibily issue. Please upgrade Photoshop. +++ Biggest update ever ahead - check news on Twitter +++ Your browser does not support HTML5 video. Source Images from Unsplash.com Four AI Connectors Gyre Best overall Stable Diffusion experience with best Outpainting, great Inpainting and CLIP Guidance. Needs local installation on Windows or Linux PC. DALL-E 2 Highest resolution and better overall image quality using text to image. It is Cloud based so Mac user can use it. Automatic1111 Popular community server. Might be useful for some special scenerious. Great Inpainting. Needs local installation. StabilityAI Cloud Fall-back if there is no other option available. It does not reach quality of the others at the moment. But we expect huge improments here in the future. Your browser does not support HTML5 video. FAQ What is required to run flying dog for DALL-E 2 & Stable Diffusion? You can use this plugin now with DALL-E 2 without installing a server (e.g. Mac users) On the server side (which can be your own Windows PC running Photoshop as well) you need to run this server here:  You can download and install this server and try some prompts in the web interface. If that runs you are ready to go with the Plugin. A decent NVidia graphics card is also required for a local installation of Gyre or Automatic1111. Photoshop  v23.5 or higher (not 23 or lower). It will run on any computer or operating system which supports the required version of Photoshop (Windows 10, MacOS, Apple Silicon, etc are all fine). For M1 Mac please activate "M1 Fix" in configuration. Which graphics card do I need for local installation? We recommend at last 6GB and it has to be from NVIDIA only, 2600 or better. 1000er series is really slow and breaks artistic workflow too much in our opinion - better use a cloud server in that case. Server part does not run on any Mac. For higher resolution than 512x512 (e.g. 1024x1024) you need more GPU memory (e.g. a 3090). What is so ​revolutionary? For the normal Photoshop user - not really into this kind AI before -in daily business the workflow will change a lot: 1. Instead of searching for stock images, just user Text-To-Image AI 2. Instead  of using Photoshop Content-Aware Fill - use Inpainting 3. If Image has to be in a spesific format/size just use Outpainting to enlarge it 4. For more artistic illustrations use Image-To-Image or specific style by using Text-To-Image 5. Instead of doing (little) correction by a brush, just use Inpainting 6. Instead of searching an object in a stock database and prepare it for adding to a scene just use Inpainting For the advanced AI user knowing all kind of AI already this is also a huge change: 1.  Instead of switching between different AI tools just use mainly this one 2. Instead of struggling with limitations of over-complicated web frontends just use the full power of Photoshop with layers, grouping, selection,... 3. Use this tool for correction of faces or hands from another AI like Midjourney V4 Can I use your plugin without having an own server? Sure get DALL-E 2 or a Dreamstudio account: https://beta.dreamstudio.ai/home That StabilityAI cloud version does not have the nice features from local server installation yet (see comparison table below).  DALL-E 2 is very powerful with its own features. Warning: StabilityAI rewrites API at the moment from GRPC to REST so it doesn't work very well right now! Update: we will publish instructions to set up your own server in cloud very soon. I am not sure about your plugin or server part. Is there a test version available? We are working on a test version right now. What about the Automatic1111 image quality? For image generation using "text to image" it offers special features like HiResFix, Tiling, or Face restoration. But it does not offer Clip Guidance and Outpainting quality is much worse than Gyre (former grpcServer). At Automatic1111 just Latent Noise will be used for Outpainting in combination with the Runway 1.5 Inpainting model. The MK2 Outpainting script of A1111 is not part of the API. Gyre on the other hand offers MK3 in combination with the Runway 1.5 Inpainting model and also Clip Guidance. Inpainting operations deliver similar results. So our recommendation which server to use is clearly Gyre but A1111 can be used additionally because the plugin supports multiple connectors. Where do I report Bugs or Feature requests? For Plugin itself use our Discord: https://discord.gg/TH9TczVP5F or our support E-Mail support@flyingdog.de We do not offer Automatic1111 installation support. What is your refund policy? Unlike our other products, flying dog for Stable Diffusion is sold directly through Adobe's marketplace. As such, it is subject to their terms and conditions, and you would need to contact Adobe Support for issues relating to the purchase itself at asupport@adobe.com (Of course we happy to provide the same technical support as always for the installed product). If the Adobe support can not help you in a refund situation, please contact our support anyway. Comparison AI Server The plugin ships with connectors to four different server. This table will be updated with each version. Things are constantly changing on a weekly basis. GYRE (former GRPCSERVER) DALL-E 2 AUTOMATIC1111 StabilityAI Cloud Price free costs per generation free costs per generation System Windows or Linux PC with NVIDA GPU none Windows or Linux PC with NVIDA GPU none Text to Image supported supported, best quality supported supported Image to Image supported not supported supported supported Inpainting supported (e.g. 1.5 Runway model), good quality supported supported (e.g. 1.5 Runway model), good quality supported (2.0 model), quality issues Speed Depends on GPU very fast Depends on GPU very fast Model SD 1.5 + RW 1.5-Inpainting + SD 2.1 + 2.0 Inpainting + Fine tuned Dreambooth models, LoRa, Depth2Image DALL-E 2 SD 1.5 +  RW1.5-Inpainting + SD 2.1+ Fine tuned Dreambooth models, LoRa, Depth2Image SD 1.5 + SD 2.1 Outpainting yes (best quality) yes (best quality) bad quality yes but quality issues Depth2Image yes no yes - just download model no Best image size 512x512 up to 1024x1024grpc 1024x1024 512x512 up to 1024x1024 512x512 up to 1024x1024 Installation 1 Click Installer not needed manual not needed Content Filter no yes no yes Mac no yes no yes flying dog Support yes yes no* yes Other Clip guidance, HiRes Fix, Tiling Variations HiRes Fix, Tiling, Face Restoration, WebUI - Internet gyre.ai OpenAI Automatic1111 on GitHub Dreamstudio * only community support on our Discord. We cannot provide full support for this kind of server because it changes very often and so it could break and not work. Another reason: the installation is more complicated. --- Gyre faq How to install Gyre locally? First you need a Windows or Linux PC with a NVidia card (2000er series or better recommened) and at least 6GB of GPU memory. Follow the instructions in this video for installation. How to install Gyre at RunPod? just follow the tutorial video here How to load models from Civitai and Huggingface? There's a bunch of models pre-configured but disabled. To enable, edit your config file and add the line: SD_ENABLE_ENGINE_1={the_id} - so for instance, to enable Realistic Vision 2.0, add SD_ENABLE_ENGINE_1=realistic-vision-v2-0 (use _2, _3, etc for more models). If you are using the docker image instead of gyre-installer, set that SD_ENABLE_ENGINE_1=.... as an environment variable. You can look in gyreconfig\extras to see the available models, and you can duplicate one of those and edit it to add other models. There's examples of different types in there (Civitai, HuggingFace, etc). It is also possible to load a model directly by config parameter. It needs a fragment. example: SD_ENGINECFG_1='!stable-diffusion-v1-5-ckpt/lyriel { "name": "lyriel", "enabled": True, "source": { "model" : "https://civitai.com/api/download/models/50127" } }' Please note that Civitai might have server issues. That might cause in download error and server needs to restarted in that case. How to change model cache path? As default model cache will be located on C: drive. As it getting more features and models the required space is more. You can change that path with XDG_CACHE_HOME={the_path} (for e.g. XDG_CACHE_HOME=e:/gyrecache) in your config file. How to activate all supported ControlNet models? add a SD_ENGINECFG_ variable with value below to your config file or use as Docker parameter. So if you have two" "SD_ENGINECFG_" already the next one would be SD_ENGINECFG_3 SD_ENGINECFG_3 ={"hintset_id": "hintset-sd15-common", "@hintset-sd15-controlnet11": [ "controlnet11/*" ] } How to activate 2.x models SD_ENABLE_ENGINE_1=stable-diffusion-v2-1 How to manually install a model locally? Each model needs to be in its own folder. Then:Example of ckpt format (ckpt needs to be only ckpt in the folder) as file "gyreconfig/extras/lyriel.yaml" --- !stable-diffusion-v1-5-ckpt/lyriel name: "lyriel" enabled: True source: local_model: "c:/models/lyriel" Or use parameter:SD_ENGINECFG_1='!stable-diffusion-v1-5-ckpt/lyriel { "name": "lyriel", "enabled": True, "source": { "local_model" : "c:/models/lyriel" } }' How to activate Swinir upscaler? The included hat-gan-x4 covers most situations better but with SD_ENABLE_ENGINE_1=swinir-x4 the Swinir-X4 can be activated in config or for Docker. Replace _1 with current var number. Can I run gyre on my AMD card? No. We recommend you use Runpod. Can I run gyre on my Mac? No. We recommend you use Runpod. --- Click here for DALL-E 2 configuration and documentation Usage of flying dog Plugin on new Gyre (former GRPCserver) (recommended)1. Start with Configuration or Installation of Server If you are using Stability AI Cloud Service just select "SD Cloud" and enter your API Key (you find in your profile in Dreamstudio) “SD Local” is using another server - check here for full documentation. For Installation of Gyre on Windows PC there is One-Click installer. A good GPU is needed from NVIDIA only (2600 or better, min 6GB RAM). You cannot run the server part on a Mac or with an AMD card. Mac computers (even the most expensive ones) have very low GPU power for this kind of task. So using a Mac as a server would be more a "proof-of-concept" than a usable solution for daily work because it is just too slow. After setup just put that link in "Gyre" Configuration. You can test the server by opening the web interface (e.g. http://127.0.0.1:5000) in the browser. In Plugin add a Gyre configuration and select a model first New: use integrated AI Studio for editing 2. Main Panel The main panel offers access to different configurations in top menu and also offers Buttons for each functions: From Text: generate an Image from a text prompt only From Image: use style-transfer for changing the whole image appearance to follow the style In-/Outpaint: Powerful editing features for adding objects in the  scene or extend the whole image Inpaint Org.: Do some little changes of an image (e.g: add sunglasses to a head) 2. Text to Image Prompt: enter whatever you want to see in text input (e.g. "a cat in suit" or just “ a duck” like in the screenshot). We have separated the so-called modifiers from main prompt input. And so you can define modifiers for individual styles below prompt input. You can also manage your own library with modifiers. Each list of modifiers can be assigned to tags which can be selected at the top as category: Switch to Modifier ideas for a quick selection of styles: So with that separation you can quickly change the style of a prompt with manually typing or copy/paste a long string again. We also implemented to comment out a line in modifier box: In this example "pastel colored" will not be part of the full prompt. Negative Prompt: add a Minus (-) in front of a line of a Modifier Steps This is the second most important parameter for Stable Diffusion AI  and we recommend that you play around with this and get some kind of feeling here with it. We made this an easy task. So you can start with a low step value first (e.g. 15) and in the preview dialog you can update individual images with higher values to get more details/better quality pictures. But this depends on the subject/prompt and also the sampler (see below). Low steps values need less computational time and results come more quickly. If you are using Cloud configuration it costs less here. So for example if you have a 3600TI. Generation of 4 images at step value 15 needs only around 8 seconds (2 seconds each). And  one image with 50 steps needs the same amount of time. Other parameters Number of images: generate between 1 and 4 images Seed: a random number where the image is connected with. Same Seed plus same other parameters results in the same image. In the preview Screen you can see the seed of each image below the image and copy it from there. Guidance Scale: higher value means that the result follows more the prompt. 7.5 is a good common value. About Sampler (advanced reading): The ancestral algorithms (Euler a and DPM2 a) do better at lower step values and Euler A tends to diverge at higher steps rather than converge. The other algorithms, especially LMS, benefit from additional steps and a slightly higher CFG. They tend to converge on an image, so as the steps go higher, the image only changes in small ways. It's the early steps where it can vary more. Euler A works better with a CFG around 5-8, and Euler/LMS do better with 7-12.  For simpler/short prompts, fewer steps, Euler 32 @ CFG 8, produces the most realistic pictures. We recommend to start with around 15 steps (in Euler A) or with 30 at LMS/DDIM and try higher values (like 50) in the preview screen for individual image updates. 3. Images Preview After starting image Generation you will see a preview window. Here you can select an Image or regenerate it with another Steps value (Update icon below image). For Image to Text and Inpainting Original a Strengths value is also provided for fine-tuning. Click On “layer Preview” for Inline-Preview of images on Canvas: Now you can browse through the images and Select the current one as well. For this preview a temporary layer group “_ai” will be generated. 4. Image to Image In this mode you can adjust an existing image to draw it in a new style. This is much more powerful than other so-called "style-transfer" algorithms but it also works differently as well because it uses the input of the prompt plus an image. On the right you see the selected image which will be used for image generation. Denoising Strength: with this parameter you can guide how strong the result should follow the original image or the prompt. So with very high value the result might be something completely different from what you might expect - just similar to Text to Image without using an input image. So we made it easy to find the correct value by offering an update of image generations in the Preview screen similar to the steps parameter for fine-tuning the result. Here again these three parameters in same example from above and result image: With this feature pretty amazing things can be made. So a children's drawing could be transferred to a more professional one or vice-versa. 5. Inpainting and Outpainting With Inpainting a part of an existing image can be changed. There are different variations of this technique known: Normal Inpainting: Place a new object in existing scene (or remove one) Original Inpainting: slightly change a part of an image Outpainting Add something new outside the image to enlarge it. 1. Normal Inpainting First remove a part of an image. Then make a selection big enough that AI can understand the surroundings and close enough to have good resolution. The erased part should match the prompt. Otherwise, the object in inpaint might be cut off at the edges. The hotel room example from the video explains quite good what is important here: If there would be only "a suitcase" the AI mostly generates not an upright one which would not fit in this area. So with additional "trolley" the results match very well to the erased area. 2. Original Inpainting If you want not an new object in scene but to just change the style of an existing part of the image. The "Original" Inpainting delivers great results. As you can see beside the prompt another image is needed - the original one. This means that the AI not only works on one image with a hole in it -  it also uses the image data from the area before it has been erased. So this technique is some kind of "cheating" because it depends on that original data. Our plugin handles it like that: the current selection of the current layer is used for the mask (image with erased area - the middle one in example above). The other layer with all image data is searched by the layer name. So  it looks for a layer with the same number in it. If it cannot be found, it uses the first layer. In configuration the Diffuser has been set to "original" here. Also set the strength value is not so high (something above 0.5) 3. Outpainting This has been added in version 1.1 (GRPC Server only) and offers state-of-the-art Outpainting. But it has the same limitations like other Outpainting (e.g. DALL-E 2) because it uses only a part of an image for extending it. So it does not know the whole context. For example: using Outpainting on existing photos is quite difficult or impossible. To keep a consistent image: try to have a strong style which delivers similar results (style, lighting,...) in Text to image as well. So for the  balloon video the style "french art poster" had been used otherwise results would not fit together here. using the same seed might also help here (advanced technique) please overlap enough pixels of original image to catch up the context details of an object at the edges might be corrupt after using it In version 1.1 only the current layer will be used from selection so flatten layers might be needed 6. Server Configuration dialog (advanced reading) Image size Stable Diffusion has been trained on images with size of 512x512. Therefore the default selection matches exakt that size. If you change that size in configuration please be very careful: only numbers which can be divided by 64 are allowed on local installation image generation might not work anymore (memory error) if these sizes are too high higher values produces quite often images with repeating content inside it On a 8GB+ GPU 768x512 or 512x768 are also good values Model In case the server provides more than one model you can click on “Get Models…” to read all available models. The Diffusion parameters of SDLocal are not needed anymore so it had been simplified at all. M1 Fix There is a bug in Photoshop which does not allow image operations while having a dialog open. So it is recommended to check this box if you are using a M1 Mac.