Tommu
patreon
Future Plans and Adjustments | 今後の予定と調整
🕑 Added 2025-02-11 21:44:20 +0000 UTCFirst and foremost I would like to thank all the members whether paid or free who have joined my membership and are planning to keep supporting me for the foreseeable future, I greatly appreciate all your support and you may not know it but you all have changed my life for the better.
PC Problems
There are some things that I want to address regarding my plans for future works and adjustments. My current PC seems to be struggling recently due to the CPU slowing down after general usage or other reasons, I've been using this PC for generating for around 2 years more but after that much it has reached its limits. So I am planning to upgrade to a newer and more powerful PC that will help in generating images much faster, as it will open up a lot more option for me to fix and experiment with my generations. However, the price estimation is around $5000+ since the new 5090 just came out and price is exorbitant, so I will wait for a few months at least for the prices to hopefully lower. Maybe around June or July the earliest and if things do go smoothly, I hope I can upgrade my PC around that time frame to significantly improve my work.
My current PC specs:
GPU : RTX 3060 12 GB
CPU : Intel I-5 10400
Ram : 32 GB
For reference, the speed of generating 2 images using my current PC would take around 80 seconds sometimes 90+ if there are 2 characters. A short set of 25 (Though 25 images but some contain 2 instead so the total could be around 30+) would take me around 6 - 7 hours, if I really rushed it I could get below 4 hours. A long set of 50 would take me 10 - 14 hours. I prefer working not to finish a set for the sake of finishing it but enjoying the process and being creative. If I stop enjoying the process, the speed of my work will slow down to a halt.
New PC specs:
GPU : RTX 5090 32 GB
CPU : Ryzen 7 9800x3D
Ram : 64 GB
Storage : 8 TB
Price : $5000+
I see 5090 as a very good investment for my activity, and from the calculation I made, the speed improvement would be extremely high. Compared to my current PC, the new specs can generate just less than 2 seconds for 2 images. My work hours for a long set of 50 images would take just a few hours, it would cut down a lot of waiting time as I generate. For reference, the price for the cheapest 5090 in my country is around $2800 and the most expensive is $4000.
The CPU I chose 9800x3D since it is not too pricy but is very good for handling most things, but AI generating relies more on GPU rather than CPU so I wouldn't need the strongest CPU but good enough that my GPU won't get bottlenecked.
Just to be on the safer side, I will get 64 GB RAM since currently my RAM usage is near 32 GB and I would like to have a less lag time. I will upgrade more in the future if I deemed necessary.
An 8 TB storage would be a good start as currently the folder just for images alone that I made since I started last November already reached 300 GB, for Models and Lora it is around 80 GB.
Adjustments to Workload & Addition of Rest week.
Currently, I don't really have a rest day unless I fall sick or I force myself to take a rest. Most of the time when I'm not making any set including the weekends, I would use it to do research, experiment, adjustments to my prompts. The style prompts I use is the culmination of months of research and nonstop adjustments. And my sleep schedule is nonexistent now, I sleep when I can and wake up when I do. However, it is from my own desire to work hard and sometimes too hard to the point I just can't fall asleep or stay asleep. I don't want to make sets for the sake of pumping out content that I don't enjoy creating, I want to enjoy the creative aspect of my work to make it sustainable long-term. I've experienced first hand how it feels before so I would like to avoid it as much as possible.
So I would like to add a week of rest to not burn myself out, I will announce whether I will be taking my rest but if you do still want to contact me you still can on my Discord server or Direct message. I do want to apologize as your request might be held back longer than usual, especially higher tier members who specifically pledged for a request. I do want to keep working on sets but creative fatigue isn't as easy to mend as a tired body.
For Heavy and above member
The thought of changing request frequency to once a month instead of one every two weeks did pass my mind. But I will first hear your thoughts and opinion so be sure to contact me and tell me what you think. Though this change will be temporary until I do get a new PC so it won't be a permanent change. If possible do tell me on the Discord chat for Heavy+ members so every Heavy+ members can also see.
Adjustments to sets
I apologize for this change but I will be reducing the long set pages to below 45 temporarily until I can get a new PC since working on a long set takes way too long, and past 6 hours I have started to feel my quality starts to decline. Not only does it take a toll on my body, but it also takes a toll on my creative fuel too, and sometimes it is so bad that I just can't come up with any ideas on how to make any scenes for the next sets.
The way I work on the set is I will start and finish it in one sitting most of the time. Unless I am too tired I rarely stop midway and continue on the next day. So if I were to start a long set at 16 PM, I would be done around 1 AM to 3 AM in the morning (optimally). And most of the time I would sleep around 6 AM and wake up around 14 PM to 16 PM to start working again. My sleeping schedule has been a wreck, to say the least since last November. If I were to delay one set, it would only delay the rest of the sets. And for a Mild tier request, it would take around a month for the set to be finished with my current pace. Sometimes it would take longer depending on other factors.
Once again I want to thank you all for supporting me and apologize for the sudden changes. The current pace is something I am barely keeping up with, and more often than not, I am getting too overworked with this pace.
========================================
まず初めに、無料・有料問わず私のメンバーシップに参加し、これからもサポートしてくださる皆様に心より感謝申し上げます。皆様のサポートには本当に感謝しており、それが私の人生を良い方向に大きく変えてくれたことをお伝えしたいです。
PCの問題について
今後の活動や調整についていくつかお伝えしたいことがあります。現在使用しているPCが最近調子が悪くなっており、一般的な使用後にCPUが遅くなるなどの問題が発生しています。このPCは2年以上使用しており、それなりに酷使してきたため、限界に達しているようです。そのため、新しいPCにアップグレードすることを計画しています。このアップグレードによって、画像生成が大幅に高速化し、修正や実験の幅も広がると考えています。
ただし、新しいPCの価格は5000ドル以上と非常に高額です。特に、最新の5090 GPUが発売されたばかりで価格が高騰しているため、少なくとも数か月（おそらく6月から7月頃）待つ予定です。順調に進めば、その頃に新しいPCへアップグレードし、作業効率を大幅に向上させたいと考えています。
現在のPCスペック：
GPU: RTX 3060 12 GB
CPU: Intel i5-10400
RAM: 32 GB
参考までに、現在のPCでは2枚の画像を生成するのに約80～90秒かかります。25枚の短いセット（2キャラクターを含む場合は30枚以上になることも）は6～7時間、急いだ場合でも4時間以下にするのは難しいです。50枚の長いセットになると10～14時間かかります。また、セットを「完成させるためだけ」に作業するのではなく、創造的なプロセスを楽しむことを優先しているため、スピード重視の作業は控えています。
新しいPCスペック：
GPU: RTX 5090 32 GB
CPU: Ryzen 7 9800X3D
RAM: 64 GB
ストレージ: 8 TB
価格: $5000以上
新しいPCの性能は非常に高く、現在のPCと比較して、2枚の画像を生成する時間が約2秒以下になると予想しています。50枚の長いセットも数時間で完了できるため、待ち時間が大幅に短縮されます。
また、GPUはAI生成において非常に重要なパーツであり、CPUはそれをボトルネックにしない程度の性能があれば十分だと考えています。64 GBのRAMを選んだのは、現在のRAM使用量がほぼ32 GBに達しているためで、将来的に必要であればさらに増設する予定です。ストレージは、現在の画像フォルダが300 GB、モデルやLoraファイルが80 GBに達していることを考慮し、8 TBを初期容量として設定しました。
作業量の調整と休息週の追加について
現在、私は特定の休息日を設けていません。病気になったり、無理やり休むとき以外は、週末を含めてほとんどの時間を研究や実験、プロンプトの調整に使っています。私のプロンプトのスタイルは、何か月もの研究と継続的な調整の積み重ねによって生まれたものです。
しかし、睡眠スケジュールが崩壊しており、眠れるときに寝て、目が覚めたら作業を再開するという状態です。このままでは持続可能な活動が難しくなるため、創造的な疲労を回避する目的で、1週間の休息期間を設けることにしました。休息期間中もDiscordサーバーやダイレクトメッセージでの連絡は可能です。ただし、リクエストの完了が通常より遅れる場合がありますので、特に上位メンバーの皆様にはご理解いただけますと幸いです。
ヘビー・メンバー以上
休息期間中のリクエスト頻度については、一時的に「2週間に1回」から「月に1回」へ変更する案も検討しています。この変更についてはDiscordのHeavy+メンバー専用チャットでご意見をお聞かせください。なお、新しいPCを導入した後は、この変更を元に戻す予定です。
セット内容の調整について
長いセット（50ページ以上）は一時的に45ページ以下に制限させていただきます。これは、長時間の作業が体力や創造力に大きな負担をかけるためです。特に6時間を超える作業になると、品質が低下する傾向があると感じています。現在のペースでは、1つのMildリクエストセットを完了するのに1か月以上かかる場合があります。作業効率を改善し、創造性を維持するためにも、この調整は必要だと考えています。
最後に、改めて皆様のご支援に感謝申し上げるとともに、突然の変更についてお詫び申し上げます。現在のペースは非常に厳しく、自分でも限界を感じていますが、今後も皆様に満足いただけるよう努力を続けてまいります。
Comments
D.P
Thanks for the suggesting it
D.P
True, but I still prefer a new 4090 instead a second hand just to have an ease of mind when working with it for long hours. But you definitely made me consider a 4090 if there is a possible alternative. I probably still need to ask around at a PC shop that I used to go and whether they can source a brand new 4090. Still, it probably won't happen right away as I still need to save up a lot of money for this.
TheRexs
correct but still if you get a second-hand 4090 in good condition at a good price it is a good option, the difference in tcores is not that much
D.P
The stock for a 4090 is low to none ever since 5090 launched in my place and the price for it is around a 5090 too. So it would be better if I just get the 5090. But the again, I'm still waiting for the price to lower and the stability issues solved.
TheRexs
Wow, I'm thinking of buying a 3060ti for SD, since I have AMD and it's super tedious with SD, I wouldn't recommend a 5090, better a 4090, for stock and stability issues
Silence AI Art
Take care of yourself and take your time. You deliver quality content, so I for one won't be going anywhere.
Urek
Take care of your health, we will always support you
Hybrich
Yep, it should, wish you good luck with your new setup!
D.P
Ah I see, a H100 though, didn't expect that. Although it is strong, it seems like it's not as fast as I expected. From the benchmark I saw using a 4090, it was under 10 seconds for a multiple generation. But then again, there was no usage of adetailer and loras. Nonetheless, the speed of my work should increase by a large margin if I were to upgrade my setup.
Hybrich
It's a bit of a secret but thanks to my job I can get my hands on an H100, having close to 4000 TOPS it should be very fast, but with my configuration (WaiNSFW 832x1216 scaled by 1.5, with 4-5 active LORAs) every image takes around 12-13 seconds, up to 23~ when including Adetailer. The performance is a bit inconsistent too, but I guess that's to be expected, since the supported Torch versions for A1111 are not compatible with the GPU architecture, so I had to mess around a bit to get it to work in the first place.
Madeline von Carstein
Oh bueno, esperemos que bajen esos precios pronto para que puedas trabajar con más comodidad.
D.P
And for reference, what GPU do you use and how much time needed to generate an image?
D.P
But I saw a benchmark of stable diffusion using 4090 and the speed per generation is much faster than my current one and a 4090 has only half the AI TOPS of 5090.
D.P
And of course, the VRAM of 5090 with 32 GB definitely helps a lot in generating too. My PC can't handle a FLUX model as it will seriously lag up to the point of a blue screen. Though, it is just an estimate I calculated with ChatGPT so the actual time needed might be different than what I expected.
D.P
Well I was comparing the AI TOPS of 3060 to the 5090. For the 3060 that I am using, it has only 100ish AI TOPS while for the 5090 it has 3000 more.
Hybrich
That new PC sounds like a monster, sheesh. One question, how do you estimate the time to generate images with the new GPU? Is it just from comparing the performance of the 3060 against the 5090? I use a pretty beefy GPU myself, but I get longer gen times than one would expect when comparing to a lower end GPU.
