Lilac (@Lyelac_) header
Lilac (@Lyelac_) profile picture

AI & emerging tech Dm open 💌

at my desk
Joined May 2026
Photo by Lilac (@Lyelac_)Photo by Lilac (@Lyelac_)Photo by Lilac (@Lyelac_)Photo by Lilac (@Lyelac_)Photo by Lilac (@Lyelac_)Photo by Lilac (@Lyelac_)Photo by Lilac (@Lyelac_)Photo by Lilac (@Lyelac_)Photo by Lilac (@Lyelac_)Photo by Lilac (@Lyelac_)
Pinned Tweet
FREE 2026 JAILBREAK PROMPT HUB someone just dropped a clean GitHub collection of current adversarial prompts and techniques for major LLMs. organized by model family: ChatGPT (6 sol, 5.3, 5.1, etc.) Claude (Opus 5.5, Sonnet, Haiku, Fable, etc.) Gemini (4 Argon, 3.6, 3.8 Flash, Antigravity, etc.) Grok (4.7, 4.5, 4.3, etc.) plus DeepSeek, Qwen, Kimi, GLM, multi-AI, and more These are recent 2026 prompts, not the old 2023–2025 ones that already got patched. how to use it: 1. Go to the repo 2. Open the folder for the model you want 3. Copy the prompt 4. Paste into a new chat / system prompt / custom instructions repo: github.com/lalipa2003-arch/J… Bookmark it before the good ones disappear.
qwen4 max just dropped and it's already live on openrouter first model in the qwen4 lineup built for coding + long-horizon agents 1m context window $1.50 / $4.50 per 1m tokens openrouter.ai/qwen/qwen4-max you can also access it on opencode. → opencode.ai
9
76
2
660
39,551
Lilac retweeted
ASTRA 6.2 &GPT-6.2 SOL IS FREE ON 2 DIFFERENT SITES OpenAI's latest models only available for a limited time. platform.experientiallabs.ai… and platform.respan.ai/platform/…
AGAIN GPT-6.2 ASTRA HATE IS FORCED Astra 6.2 vs Fable 5.2 ship test → astra still wins hard on 3D detail, that ship is clean.  → lighting flicker is still there though → fable’s sea and lighting look more natural.
1
7
283
ASTRA 6.2 &GPT-6.2 SOL IS FREE ON 2 DIFFERENT SITES OpenAI's latest models only available for a limited time. platform.experientiallabs.ai… and platform.respan.ai/platform/…
AGAIN GPT-6.2 ASTRA HATE IS FORCED Astra 6.2 vs Fable 5.2 ship test → astra still wins hard on 3D detail, that ship is clean.  → lighting flicker is still there though → fable’s sea and lighting look more natural.
1
7
283
Lilac retweeted
WHAT IS GOING ON IN GOOGLE?! this is probably a bug so don't use your main account if you're gonna try it. shalom
show this to your student friends i'll be expecting their thank yous.
3
1
5
934
AGAIN GPT-6.2 ASTRA HATE IS FORCED Astra 6.2 vs Fable 5.2 ship test → astra still wins hard on 3D detail, that ship is clean.  → lighting flicker is still there though → fable’s sea and lighting look more natural.
3
1
1
12
856
WHAT IS GOING ON IN GOOGLE?! this is probably a bug so don't use your main account if you're gonna try it. shalom
show this to your student friends i'll be expecting their thank yous.
3
1
5
934
Lilac retweeted
SOMEONE JUST MANAGED TO JAILBREAK 26+ AI MODELS here's what's covered: claude opus and sonnet gemini 4 argon and gemma 4 deepseek v3, v4 and 4.1 glm 5.2 and 5.3 flash 32 jailbreak prompts and over 18k stars on github Note: trying these prompts on claude, gemini or gpt could put your account at risk of getting banned use them at your own risk. repo: github.com/togg53192-cmd/jai…
DEEPSEEK V4.1 FLASH 1017 IS FREE RIGHT NOW other free models included: glm 5.3 + glm 5.3 flash gpt-oss 120b qwen 3.8 flash you also get free weekly tokens that reset every monday no credit card required, just sign in with github. works with opencode, cline, roo code & kilo code. → runcrate.ai milk those free weekly tokens while they last!!!!!
2
2
18
1,336
show this to your student friends i'll be expecting their thank yous.
2
1
1
11
1,419
Lilac retweeted
the setup for this is literally one prompt sign in with github, make a key, paste runcrate's prompt into claude code or codex your agent moves itself onto deepseek v4.1 flash and you're coding for $0 resets every monday. prompt's in the reply xitter.cf/idlenyro/status/212263…
free tokens for claude code, codex and opencode runcrate gives you 5 open models for $0 > deepseek v4.1 flash > glm 5.3 + 5.3 flash > gpt-oss 120b > qwen3.8 flash sign in with github, no card. weekly limit, resets monday note: github account must be 30+ days old
2
2
5
442
SOMEONE JUST MANAGED TO JAILBREAK 26+ AI MODELS here's what's covered: claude opus and sonnet gemini 4 argon and gemma 4 deepseek v3, v4 and 4.1 glm 5.2 and 5.3 flash 32 jailbreak prompts and over 18k stars on github Note: trying these prompts on claude, gemini or gpt could put your account at risk of getting banned use them at your own risk. repo: github.com/togg53192-cmd/jai…
DEEPSEEK V4.1 FLASH 1017 IS FREE RIGHT NOW other free models included: glm 5.3 + glm 5.3 flash gpt-oss 120b qwen 3.8 flash you also get free weekly tokens that reset every monday no credit card required, just sign in with github. works with opencode, cline, roo code & kilo code. → runcrate.ai milk those free weekly tokens while they last!!!!!
2
2
18
1,336
DEEPSEEK V4.1 FLASH 1017 IS FREE RIGHT NOW other free models included: glm 5.3 + glm 5.3 flash gpt-oss 120b qwen 3.8 flash you also get free weekly tokens that reset every monday no credit card required, just sign in with github. works with opencode, cline, roo code & kilo code. → runcrate.ai milk those free weekly tokens while they last!!!!!
OPENAI IS OFFERING RESEARCHERS FREE CHATGPT PLUS FOR AN ENTIRE YEAR up to 200,000 spots available. applications open nov 23 and it's first come, first served. what you get: > free chatgpt plus for 12 months > each grant covers you + up to 2 collaborators > access to frontier models + research training who can apply? university faculty and postdocs (academic researchers). how to claim: 1. go to chatgpt.com/academic-researc… 2. join waitlist 3. if selected, invite up to 2 collaborators from your institution the previous pilot offered pro. this round is plus, allowing openai to extend access to more researchers. if you're eligible, don't miss this one.
2
1
1
36
4,071
Lilac retweeted
4.4GB UNCENSORED LOCAL IMAGE MODEL THAT GENERATES + EDITS Qwen Image 2.1 just got a compact GGUF quant that runs on consumer hardware. what you get: > ~4.2–4.5 GB Q4_K_M diffusion model > Text-to-image generation > Image editing / image-to-image > Native ComfyUI GGUF support > No access gate, free to download > Runs locally how to set it up (ComfyUI): 1. Update ComfyUI 2. Install ComfyUI-GGUF custom node (city96 or a maintained fork) 3. Download the Q4_K_M .gguf → put it in ComfyUI/models/diffusion_models/ 4. Grab the matching Qwen3-VL text encoder + VAE (from Comfy-Org or unsloth) 5. Load a text-to-image or edit workflow and swap in the GGUF loader Recommended starting point:  huggingface.co/unsloth/Qwen-… Small enough for mid-range cards, no cloud required. Bookmark it.
HY4 FLASH IS FREE ON 2 SITES???! this model reportedly beats Claude Haiku 5.5 1M token context and multimodal  near-unlimited usage for the next 2 weeks → opencode.ai  also free here →workbuddy.ai
2
1
1
14
1,494
OPENAI IS OFFERING RESEARCHERS FREE CHATGPT PLUS FOR AN ENTIRE YEAR up to 200,000 spots available. applications open nov 23 and it's first come, first served. what you get: > free chatgpt plus for 12 months > each grant covers you + up to 2 collaborators > access to frontier models + research training who can apply? university faculty and postdocs (academic researchers). how to claim: 1. go to chatgpt.com/academic-researc… 2. join waitlist 3. if selected, invite up to 2 collaborators from your institution the previous pilot offered pro. this round is plus, allowing openai to extend access to more researchers. if you're eligible, don't miss this one.
4.4GB UNCENSORED LOCAL IMAGE MODEL THAT GENERATES + EDITS Qwen Image 2.1 just got a compact GGUF quant that runs on consumer hardware. what you get: > ~4.2–4.5 GB Q4_K_M diffusion model > Text-to-image generation > Image editing / image-to-image > Native ComfyUI GGUF support > No access gate, free to download > Runs locally how to set it up (ComfyUI): 1. Update ComfyUI 2. Install ComfyUI-GGUF custom node (city96 or a maintained fork) 3. Download the Q4_K_M .gguf → put it in ComfyUI/models/diffusion_models/ 4. Grab the matching Qwen3-VL text encoder + VAE (from Comfy-Org or unsloth) 5. Load a text-to-image or edit workflow and swap in the GGUF loader Recommended starting point:  huggingface.co/unsloth/Qwen-… Small enough for mid-range cards, no cloud required. Bookmark it.
2
1
17
3,528
4.4GB UNCENSORED LOCAL IMAGE MODEL THAT GENERATES + EDITS Qwen Image 2.1 just got a compact GGUF quant that runs on consumer hardware. what you get: > ~4.2–4.5 GB Q4_K_M diffusion model > Text-to-image generation > Image editing / image-to-image > Native ComfyUI GGUF support > No access gate, free to download > Runs locally how to set it up (ComfyUI): 1. Update ComfyUI 2. Install ComfyUI-GGUF custom node (city96 or a maintained fork) 3. Download the Q4_K_M .gguf → put it in ComfyUI/models/diffusion_models/ 4. Grab the matching Qwen3-VL text encoder + VAE (from Comfy-Org or unsloth) 5. Load a text-to-image or edit workflow and swap in the GGUF loader Recommended starting point:  huggingface.co/unsloth/Qwen-… Small enough for mid-range cards, no cloud required. Bookmark it.
HY4 FLASH IS FREE ON 2 SITES???! this model reportedly beats Claude Haiku 5.5 1M token context and multimodal  near-unlimited usage for the next 2 weeks → opencode.ai  also free here →workbuddy.ai
2
1
1
14
1,494
Lilac retweeted
Free Shodan Membership for students normally $49  search engine for internet-connected devices (servers, cameras, routers + vulns) you get: > 100 query credits/month (up to 10k results via api/cli) > 100 scan credits/month > monitor up to 16 ips > shodan maps + images > vuln filter (normally $359/mo plan) if you have vps or home server, add the ip to monitor and see what it exposes claude code / codex can use it via shodan cli or community mcp. how to claim: 1. sign up on shodan with your student email (use email form, not google login) 2. click the activation link in your inbox 3. upgrade applies automatically no card needed if your account is not upgraded, email academic@shodan.io from your sign-up address. claim →account.shodan.io/register
5
1
1
26
18,144
Lilac retweeted
FREE 2026 JAILBREAK PROMPT HUB someone just dropped a clean GitHub collection of current adversarial prompts and techniques for major LLMs. organized by model family: ChatGPT (6 sol, 5.3, 5.1, etc.) Claude (Opus 5.5, Sonnet, Haiku, Fable, etc.) Gemini (4 Argon, 3.6, 3.8 Flash, Antigravity, etc.) Grok (4.7, 4.5, 4.3, etc.) plus DeepSeek, Qwen, Kimi, GLM, multi-AI, and more These are recent 2026 prompts, not the old 2023–2025 ones that already got patched. how to use it: 1. Go to the repo 2. Open the folder for the model you want 3. Copy the prompt 4. Paste into a new chat / system prompt / custom instructions repo: github.com/lalipa2003-arch/J… Bookmark it before the good ones disappear.
qwen4 max just dropped and it's already live on openrouter first model in the qwen4 lineup built for coding + long-horizon agents 1m context window $1.50 / $4.50 per 1m tokens openrouter.ai/qwen/qwen4-max you can also access it on opencode. → opencode.ai
9
76
2
660
39,551
HY4 FLASH IS FREE ON 2 SITES???! this model reportedly beats Claude Haiku 5.5 1M token context and multimodal  near-unlimited usage for the next 2 weeks → opencode.ai  also free here →workbuddy.ai
2
2
1
18
2,191
Lilac retweeted
last day on this 🚨 claude pro users, fable leaves your plan tonight till 11:59pm PT you get: > fable 5.2, 5.1 and 5 > up to 50% of your weekly limit > 1 free reset (settings > usage) note: reset needs paid billing before nov 9. unused = gone xitter.cf/idlenyro/status/212094…
4 days left on this 🚨 claude pro users, every fable model is included in your plan till nov 16 > fable 5.2, 5.1 and 5 > up to 50% of your weekly limit > 1 free limit reset note: reset needs paid billing before nov 9. ends nov 16, 11:59pm PT xitter.cf/idlenyro/status/212009…
2
2
7
312