• úvod
  • témata
  • události
  • tržiště
  • diskuze
  • nástěnka
  • přihlásit
    registrace
    ztracené heslo?
    KERRAYAI Art - Tvorba za pomocí umělé inteligence: Midjourney, DALL·E 2, Stable Diffusion, OpenAI
    HITOMI
    HITOMI --- ---
    #nástroj #stablediffusion #text #návod

    Tohle by mohlo zachraňovat životy :) Návod, jak si v SD rozdělit obrázek na části pro snazší promptování s více objekty :)
    Regional Prompter: Control image composition in Stable Diffusion - Stable Diffusion Art
    https://stable-diffusion-art.com/regional-prompter/
    KERRAY
    KERRAY --- ---
    #nástroj tipy na trénování vlastních LoRA modelů
    Reddit - Dive into anything
    https://www.reddit.com/r/StableDiffusion/comments/13dh7ql/after_training_50_lora_models_here_is_what_i/

    Style Training :

    - use 30-100 images (avoid same subject, avoid big difference in style)

    - good captioning (better caption manually instead of BLIP) with alphanumeric trigger words (styl3name).

    - use pre-existing style keywords (i.e. comic, icon, sketch)

    - caption formula styl3name, comic, a woman in white dress

    - train with a model that can already produce a close looking style that you are trying to acheive.

    - avoid stablediffusion base model beacause it is too diverse and we want to remain specific


    Person/Character Training:

    - use 30-100 images (atleast 20 closeups and 10 body shots)

    - face from different angles, body in different clothing and in different lighting but not too much diffrence, avoid pics with eye makeup

    - good captioning (better caption manually instead of BLIP) with alphanumeric trigger words (ch9ractername)

    - avoid deep captioning like "a 25 year woman in pink printed tshirt and blue ripped denim striped jeans, gold earing, ruby necklace"

    - caption formula ch9ractername, a woman in pink tshirt and blue jeans

    - for real person, train on RealisticVision model, Lora trained on RealisticVision works with most of the models

    - for character training use train with a model that can already produce a close looking character (i.e. for anime i will prefer anythinv3)

    - avoid stablediffusion base model beacause it is too diverse and we want to remain specific
    KERRAY
    KERRAY --- ---
    #video #obrázek
    Lord of the Rings by Wes Anderson Trailer | The Whimsical Fellowship
    https://youtu.be/KrjL_TSOFrI

    • MidJourney (generated images subject matter, backgrounds)

    • Stable Diffusion for upscaling(?)

    • ElevenLabs (AI text-to-voice/voice cloning) Narrator

    • D-Id (AI mouth/eye/head movements + lip syncing)

    • ChatGPT(4) For the story/dialogue (which you insert into ElevenLabs)

    • Adobe (Premiere Pro/After Effects) video editing/typography
    QWWERTY
    QWWERTY --- ---
    zahlednuto v dnesnim Midjourney channelu :))
    uz i ty ruce vypadaji docela dobre

    QWWERTY
    QWWERTY --- ---
    Voldy Retard GUide to Stable Diffusion

    --GUIDE--
    https://rentry.org/voldy
    DAVE2
    DAVE2 --- ---
    GREGI
    GREGI --- ---
    SATAI: brdavky povoleny je to v popisku!:D
    SATAI
    SATAI --- ---
    GREGI: vyberu něco bezbradavkovatého ;-)
    GREGI
    GREGI --- ---
    SATAI: muzeme udelat i porn one vid:)
    SATAI
    SATAI --- ---
    GREGI: a já mám vše jen z pornpen.ai :-/
    GREGI
    GREGI --- ---
    Myslim ze cista galerka by byla fajn tak kdyztak hlasujte:)
    [GREGI @ nyx :: zadosti o zalozeni klubu - vsechny pohromade]
    NAVARA
    NAVARA --- ---
    GREGI: Zapomeň na léčbu rakoviny např
    GREGI
    GREGI --- ---
    DAVE2: Autorske jeste tak nejak chapu, ale opravdu zrusit patenty? Chapu tu utopistickou myslenku volneho proudu znalosti a inovaci. Nebyla by to ale pak pro spoustu projektu velka rana v ramci financovani od investoru, kteri by logicky nemeli zadnou motovaci do projektu nalejt cash, tzn obvzlast ty drahe projekty napr. v ramci energetiky by nemely moc sanci cokoliv dat dohromady?
    BUBBLE
    BUBBLE --- ---
    The following video demonstrates the use of Inworld AI to power NPCs in Skyrim with a custom-built story engine for dialogue interactions.

    With Inworld's powerful AI character platform, you can now chat with NPCs in Skyrim on your own terms. Get almost instant responses that stay true to each character's unique personality, and immerse yourself even further in the game's universe. This is the future of roleplaying games.

    Watch the video of me wandering around the Whiterun, talking with Lydia about what she does while she is waiting, or asking secret love of Jon Battle-Born, or finding out facts about skooma or learning the actual personality and backstory of our glorious memeguard, "I used to be an adventurer like you, then I took an arrow in the knee"!

    Future of Elder Scrolls & RPG - Skyrim NPCs & Inworld AI (like GPT-4 for gaming)
    https://www.youtube.com/watch?v=d6sVWEu9HWU&ab_channel=Bloc
    ARAON
    ARAON --- ---
    New Blizzard patent details an AI system to make art for its games | PC Gamer
    https://www.pcgamer.com/new-blizzard-patent-details-an-ai-system-to-make-art-for-its-games/
    M4JKL05
    M4JKL05 --- ---
    NAGASAWA: Saru Conner by mohla v ČR alternovat Lucka Bílá a Kamil Střihavka, T-800 Vémola
    KERRAY
    KERRAY --- ---
    #video
    https://twitter.com/ChristianF369/status/1651607149804498946
    LOJZA
    LOJZA --- ---
    KERRAY: Paráda, to chci vidět!
    NAGASAWA
    NAGASAWA --- ---
    Terminator: The Musical
    https://youtu.be/BUM6UN__GK4
    KERRAY
    KERRAY --- ---
    #text #nástroj užitečné prompty

    Reddit - Dive into anything
    https://www.reddit.com/r/ChatGPT/comments/12zrigy/better_than_prompt_engineering_just_using_these_6/

    I’ve found myself using the same snippets over and over again to give chatgpt that bit of guidance, so I thought I’d share them here to see if it’s helpful, and to ask whether you’ve found similar helpful snippets.

    “Let's think step by step”

    Just adding this at the end of your problem is probably one of the best-proven ‘tricks’. It was already known that it improved GPT-3 math problem solving skills by 59%, and although GPT-4 is already much smarter by itself this still helps - it works well even with very abstract problems, example:

    I have a background in engineering, coding, solo-entrepreneurship, language learning and cognitive science. What is a helpful product I can build in the AI space? Let's think step by step


    Reflection/Evaluation:

    - “Identify any unclear or ambiguous information in your response, and rephrase it for clarity.”
    - “Try to argue against your own output and see if you can find any flaws. If so, address them. Walk me through the process”
    - “Rate on a scale from 0-5 how well you satisfied each point in the initial prompt. Be very critical, no need to justify yourself. If 3 or lower, rewrite to address.” (this is a hit or miss btw - I’ve seen instances where ChatGPT just rates itself 5/5 all the time)

    All of these snippets do similar things, forcing chatgpt to reflect on and evaluate its previous output, so that it can find flaws and improve. This kind of reflective technique in general (not these specific prompts) have been shown to improve GPT-4 accuracy by 30%

    “What additional input do you need from me to help you write better output?”

    This basically asks ChatGPT to evaluate its own output, as well as the initial prompt, and suggest additional information to improve the output. We often know we can get better results by providing more detailed input, but this prompt really helps to know which input exactly would be most helpful to provide.

    “Please summarise what I am asking for you before you begin your answer.”

    This works especially well for complicated problems such as coding. By asking ChatGPT to summarise the problem itself, it not only helps itself come up with a more accurate solution, but you can also check and correct ChatGPT’s understanding.
    Kliknutím sem můžete změnit nastavení reklam